We propose a novel self-supervised Video Object Segmentation (VOS) approach that strives to achieve better visual correspondence across frames and object-background discriminability for accurate object segmentation. Distinct from previous self-supervised VOS methods, our approach is based on a learning loss formulation that takes into account both object and background information to ensure object-background discriminability, rather than using only object appearance. The objective function comprises cutout-based reconstruction (cutout region represents part of a frame, whose pixels are replaced with some constant values) and tag prediction loss terms. The cutout-based reconstruction term utilizes a simple cutout scheme to learn the pixel-wise correspondence between the current and previous frames in order to reconstruct the original current frame with added cutout region in it. The introduced cutout patch guides the model to focus on the reappearance of scene parts , thereby implicitly equipping the model to address occlusion-based scenarios Next, the tag prediction term encourages object-background separability by grouping tags of all pixels in the cutout region that are similar, while separating them from the tags of the rest of the reconstructed pixels. Additionally, we introduce a zoom-in scheme that addresses the problem of small object segmentation by capturing fine structural information at multiple scales. Our proposed approach, termed CT-VOS, achieves state-of-the-art results on two challenging benchmarks: DAVIS-2017 and Youtube-VOS. A detailed ablation showcases the importance of the proposed loss formulation to effectively establish correspondences, object-background discriminability, and the impact of our zoom-in scheme to accurately segment small-sized objects.

Anomaly detection has recently gained increasing attention in the field of computer vision, likely due to its broad set of applications ranging from product fault detection on industrial production lines and impending event detection in video surveillance to finding lesions in medical scans. Regardless of the domain, anomaly detection is typically framed as a one-class classification task, where the learning is conducted on normal examples only. An entire family of successful anomaly detection methods is based on learning to reconstruct masked normal inputs (e.g. patches, future frames, etc.) and exerting the magnitude of the reconstruction error as an indicator for the abnormality level. Unlike other reconstruction-based methods, we present a novel self-supervised masked convolutional transformer block (SSMCTB) that comprises the reconstruction-based functionality at a core architectural level. The proposed self-supervised block is extremely flexible, enabling information masking at any layer of a neural network and being compatible with a wide range of neural architectures. In this work, we extend our previous self-supervised predictive convolutional attentive block (SSPCAB) with a 3D masked convolutional layer, a transformer for channel-wise attention, as well as a novel self-supervised objective based on Huber loss. Furthermore, we show that our block is applicable to a wider variety of tasks, adding anomaly detection in medical images and thermal videos to the previously considered tasks based on RGB images and surveillance videos. We exhibit the generality and flexibility of SSMCTB by integrating it into multiple state-of-the-art neural models for anomaly detection, bringing forth empirical results that confirm considerable performance improvements on five benchmarks: MVTec AD, BRATS, Avenue, ShanghaiTech, and Thermal Rare Event.

The field of visual object tracking is dominated by methods that combine simple tracking algorithms and ad hoc schemes. Probabilistic tracking algorithms, which are leading in other fields, are surprisingly absent from the leaderboards. We found that accounting for distance in target kinematics, exploiting detector confidence and modelling non-uniform clutter characteristics is critical for a probabilistic tracker to work in visual tracking. Previous probabilistic methods fail to address most or all these aspects, which we believe is why they fall so far behind current state-of-the-art (SOTA) methods (there are no probabilistic trackers in the MOT17 top 100). To rekindle progress among probabilistic approaches, we propose a set of pragmatic models addressing these challenges, and demonstrate how they can be incorporated into a probabilistic framework. We present BASE (Bayesian Approximation Single-hypothesis Estimator), a simple, performant and easily extendible visual tracker, achieving state-of-the-art (SOTA) on MOT17 and MOT20, without using Re-Id. Code available at https://github.com/ffi-no/paper-base-visapp-2024.

Taking a better look at subjects of interest helps humans to improve confidence in their age estimation. Unlike still images, sequences offer spatio-temporal dynamic information that contains many cues related to age progression. A review of previous work on video-based age estimation indicates that this is an underexplored field of research. This may be caused by a lack of well-defined and publicly accessible video benchmark protocol, as well as the absence of video-oriented training data. To address the former issue, we propose a carefully designed video age estimation benchmark protocol and make it publicly available. To address the latter issue, we design a video-specific age estimation method that leverages pseudo-labeling and semi-supervised learning. Our results show that the proposed method outperforms image-based baselines on both offline and online benchmark protocols, while the online estimation stability is improved by more than 50%.

Deep learning has shown remarkable success in remote sensing change detection (CD), aiming to identify semantic change regions between co-registered satellite image pairs acquired at distinct time stamps. However, existing convolutional neural network (CNN) and transformer-based frameworks often struggle to accurately segment semantic change regions. Moreover, transformers-based methods with standard self-attention suffer from quadratic computational complexity with respect to the image resolution, making them less practical for CD tasks with limited training data. To address these issues, we propose an efficient change detection (CD) framework, efficient global and local context aggregation network (ELGC-Net), which leverages rich contextual information to precisely estimate change regions while reducing the model size. Our ELGC-Net comprises a Siamese encoder, fusion modules, and a decoder. The focus of our design is the introduction of an efficient local-global context aggregator (ELGCA) module within the encoder, capturing enhanced global context and local spatial information through a novel pooled-transpose (PT) attention and depthwise convolution, respectively. The PT attention employs pooling operations for robust feature extraction and minimizes computational cost with transposed attention. Extensive experiments on three challenging CD datasets demonstrate that ELGC-Net outperforms existing methods. Compared to the recent transformer-based CD approach (ChangeFormer), ELGC-Net achieves a 1.4% gain in intersection over union (IoU) metric on the LEVIR-CD dataset, while significantly reducing trainable parameters. Our proposed ELGC-Net sets a new state-of-the-art (SOTA) performance in remote sensing CD benchmarks. Finally, we also introduce ELGC-Net-LW, a lighter variant with significantly reduced computational complexity, suitable for resource-constrained settings, while achieving comparable performance.

A self-supervised multi-task learning (SSMTL) framework for video anomaly detection was recently introduced in literature. Due to its highly accurate results, the method attracted the attention of many researchers. In this work, we revisit the self-supervised multi-task learning framework, proposing several updates to the original method. First, we study various detection methods, e.g. based on detecting high-motion regions using optical flow or background subtraction, since we believe the currently used pre-trained YOLOv3 is suboptimal, e.g. objects in motion or objects from unknown classes are never detected. Second, we modernize the 3D convolutional backbone by introducing multi-head self-attention modules, inspired by the recent success of vision transformers. As such, we alternatively introduce both 2D and 3D convolutional vision transformer (CvT) blocks. Third, in our attempt to further improve the model, we study additional self-supervised learning tasks, such as predicting segmentation maps through knowledge distillation, solving jigsaw puzzles, estimating body pose through knowledge distillation, predicting masked regions (inpainting), and adversarial learning with pseudo-anomalies. We conduct experiments to assess the performance impact of the introduced changes. Upon finding more promising configurations of the framework, dubbed SSMTL++v1 and SSMTL++v2, we extend our preliminary experiments to more data sets, demonstrating that our performance gains are consistent across all data sets. In most cases, our results on Avenue, ShanghaiTech and UBnormal raise the state-of-the-art performance bar to a new level.

Class-Incremental Learning is a challenging problem in machine learning that aims to extend previously trained neural networks with new classes. This is especially useful if the system is able to classify new objects despite the original training data being unavailable. Although the semantic segmentation problem has received less attention than classification, it poses distinct problems and challenges, since previous and future target classes can be unlabeled in the images of a single increment. In this case, the background, past and future classes are correlated and there exists a background-shift.

In this paper, we address the problem of how to model unlabeled classes while avoiding spurious feature clustering of future uncorrelated classes. We propose to use Evidential Deep Learning to model the evidence of the classes as a Dirichlet distribution. Our method factorizes the problem into a separate foreground class probability, calculated by the expected value of the Dirichlet distribution, and an unknown class (background) probability corresponding to the uncertainty of the estimate. In our novel formulation, the background probability is implicitly modeled, avoiding the feature space clustering that comes from forcing the model to output a high background score for pixels that are not labeled as objects. Experiments on the incremental Pascal VOC and ADE20k benchmarks show that our method is superior to the state of the art, especially when repeatedly learning new classes with increasing number of increments.

We propose a novel approach to translate unpaired contrast computed tomography (CT) scans to noncontrast CT scans and the other way around. Solving this task has two important applications: (i) to automatically generate contrast CT scans for patients for whom injecting contrast substance is not an option, and (ii) to enhance the alignment between contrast and non-contrast CT by reducing the differences induced by the contrast substance before registration.Our approach is based on cycle-consistent generative adversarial convolutional transformers, for short, CyTran. Our neural model can be trained on unpaired images, due to the integration of a multi-level cycleconsistency loss. Aside from the standard cycle-consistency loss applied at the image level, we propose to apply additional cycle-consistency losses between intermediate feature representations, which enforces the model to be cycle-consistent at multiple representations levels, leading to superior results. To deal with high-resolution images, we design a hybrid architecture based on convolutional and multi-head attention layers. In addition, we introduce a novel data set, Coltea-Lung-CT-100W, containing 100 3D triphasic lung CT scans (with a total of 37,290 images) collected from 100 female patients (there is one examination per patient). Each scan contains three phases (non-contrast, early portal venous, and late arterial), allowing us to perform experiments to compare our novel approach with state-of-the-art methods for image style transfer.Our empirical results show that CyTran outperforms all competing methods. Moreover, we show that CyTran can be employed as a preliminary step to improve a state-of-the-art medical image alignment method. We release our novel model and data set as open source at: https://github.com/ristea/cycletransformer.Our qualitative and subjective human evaluations reveal that CyTran is the only approach that does not introduce visual artifacts during the translation process. We believe this is a key advantage in our application domain, where medical images need to precisely represent the scanned body parts. (c) 2023 Elsevier B.V. All rights reserved.

The choice of the material base to which the material decomposition is performed in dual-energy computed tomography may affect the quality of reconstructed images. Resulting inaccuracies may lower their diagnostic value, or if the data are used for radiation treatment planning, the accuracy of such plans. The aim of this work is to investigate how the commonly used (water, bone) (WB), (water, iodine) (WI), and (approximate photoelectric effect, Compton scattering) (PC) doublets affect the reconstructed linear attenuation coefficient in the case of the Alvarez–Macovski (AM) method. The performance of this method is also compared to the performance of the dual-energy iterative reconstruction algorithm DIRA. In both cases, the study is performed using simulations.

The results show that the PC and WB doublets accurately predicted the linear attenuation coefficient (LAC) values for human tissues and elements with Z = 1, …, 20, in the 20–150 keV range, though there was a small (<5% discrepancy in the 20–35 keV range. The WI doublet did not represent the tissues as well as PC and WB; the largest discrepancies (>50% in some cases) were in the 20–40 keV range.

LACs reconstructed with the AM and DIRA followed this trend. AM produced artifacts when iodine was present in the phantom together with human tissues since AM can only work with one doublet at a time. It was shown that these artifacts could be avoided with DIRA using different doublets at different spatial positions, i.e., WB for soft and bone tissue and WI for the iodine solution.

Super-resolving medical images can help physicians in providing more accurate diagnostics. In many situations, computed tomography (CT) or magnetic resonance imaging (MRI) techniques capture several scans (modes) during a single investigation, which can jointly be used (in a multimodal fashion) to further boost the quality of super-resolution results. To this end, we propose a novel multimodal multi-head convolutional attention module to super-resolve CT and MRI scans. Our attention module uses the convolution operation to perform joint spatial-channel attention on multiple concatenated input tensors, where the kernel (receptive field) size controls the reduction rate of the spatial attention, and the number of convolutional filters controls the reduction rate of the channel attention, respectively. We introduce multiple attention heads, each head having a distinct receptive field size corresponding to a particular reduction rate for the spatial attention. We integrate our multimodal multi-head convolutional attention (MMHCA) into two deep neural architectures for super-resolution and conduct experiments on three data sets. Our empirical results show the superiority of our attention module over the state-of-the-art attention mechanisms used in super-resolution. Moreover, we conduct an ablation study to assess the impact of the components involved in our attention module, e.g. the number of inputs or the number of heads. Our code is freely available at https://github.com/lilygeorgescu/MHCA.

Accurate and robust visual object tracking is one of the most challenging and fundamental computer vision problems. It entails estimating the trajectory of the target in an image sequence, given only its initial location, and segmentation, or its rough approximation in the form of a bounding box. Discriminative Correlation Filters (DCFs) and deep Siamese Networks (SNs) have emerged as dominating tracking paradigms, which have led to significant progress. Following the rapid evolution of visual object tracking in the last decade, this survey presents a systematic and thorough review of more than 90 DCFs and Siamese trackers, based on results in nine tracking benchmarks. First, we present the background theory of both the DCF and Siamese tracking core formulations. Then, we distinguish and comprehensively review the shared as well as specific open research challenges in both these tracking paradigms. Furthermore, we thoroughly analyze the performance of DCF and Siamese trackers on nine benchmarks, covering different experimental aspects of visual tracking: datasets, evaluation metrics, performance, and speed comparisons. We finish the survey by presenting recommendations and suggestions for distinguished open challenges based on our analysis.

The present work proposes the use of point clouds differential entropy as a method for reverse engineering quality assessment. This quality assessment can be used to measure the deviation of objects made with additive manufacturing or CNC techniques. The quality of the execution is intended as a measure of the deviation of the geometry of the obtained object compared to the original CAD. This paper proposes the use of the quality index of the CorAl method to assess the quality of an objects compared to its original CAD. This index, based on the differential entropy, takes on a value the closer to 0 the more they obtained object is close to the original geometry. The advantage of this method is to have a global synthetic index. It is however possible to have entropy maps of the individual points to verify which are the areas with the greatest deviation. The method is robust for comparing point clouds at different densities. Objects obtained by additive manufacturing with different print qualities were used. The quality index evaluated for each object, as defined in the CorAl method, turns out to be gradually closer to 0 as the quality of the piece's construction increases.

Video instance segmentation is one of the core problems in computer vision. Formulating a purely learning-based method, which models the generic track management required to solve the video instance segmentation task, is a highly challenging problem. In this work, we propose a novel learning framework where the entire video instance segmentation problem is modeled jointly. To this end, we design a graph neural network that in each frame jointly processes all detections and a memory of previously seen tracks. Past information is considered and processed via a recurrent connection. We demonstrate the effectiveness of the proposed approach in comprehensive experiments. Our approach operates online at over 25 FPS and obtains 16.3 AP on the challenging OVIS benchmark, setting a new state-of-the-art. We further conduct detailed ablative experiments that validate the different aspects of our approach. Code is available at https://github.com/emibr948/RGNNVIS-PlusPlus.

Predicting the motion of other road agents enables autonomous vehicles to perform safe and efficient path planning. This task is very complex, as the behaviour of road agents depends on many factors and the number of possible future trajectories can be considerable (multi-modal). Most prior approaches proposed to address multi-modal motion prediction are based on complex machine learning systems that have limited interpretability. Moreover, the metrics used in current benchmarks do not evaluate all aspects of the problem, such as the diversity and admissibility of the output. The authors aim to advance towards the design of trustworthy motion prediction systems, based on some of the requirements for the design of Trustworthy Artificial Intelligence. The focus is on evaluation criteria, robustness, and interpretability of outputs. First, the evaluation metrics are comprehensively analysed, the main gaps of current benchmarks are identified, and a new holistic evaluation framework is proposed. Then, a method for the assessment of spatial and temporal robustness is introduced by simulating noise in the perception system. To enhance the interpretability of the outputs and generate more balanced results in the proposed evaluation framework, an intent prediction layer that can be attached to multi-modal motion prediction models is proposed. The effectiveness of this approach is assessed through a survey that explores different elements in the visualisation of the multi-modal trajectories and intentions. The proposed approach and findings make a significant contribution to the development of trustworthy motion prediction systems for autonomous vehicles, advancing the field towards greater safety and reliability.

On a shutter press, modern handheld cameras capture multiple images in rapid succession and merge them to generate a single image. However, individual frames in a burst are misaligned due to inevitable motions and contain multiple degradations. The challenge is to properly align the successive image shots and merge their complimentary information to achieve high-quality outputs. Towards this direction, we propose Burstormer: a novel transformer-based architecture for burst image restoration and enhancement. In comparison to existing works, our approach exploits multi-scale local and non-local features to achieve improved alignment and feature fusion. Our key idea is to enable inter-frame communication in the burst neighborhoodsf or information aggregation and progressive fusion while modeling the burst-wide context. However, the input burst frames need to be properly aligned before fusing their information. Therefore, we propose an enhanced deformable alignment module for aligning burst features with regards to the reference frame. Unlike existing methods, the proposed alignment module not only aligns burst features but also exchanges feature information and maintains focused communication with the reference frame through the proposed reference-based feature enrichment mechanism, which facilitates handling complex motions. Aft er multi-level alignment and enrichment, we re-emphasize on inter-frame communication within burst using a cyclic burst sampling module. Finally, the inter-frame information is aggregated using the proposed burst feature fusion module followed by progressive upsampling. Our Burstormer outperforms state-of-the-art methods on burst super-resolution, burst denoising and burst low-light enhancement. Our codes and pre-trained models are available at https://github.com/akshaydudhane16/Burstormer.

Large-scale multi-modal training with image-text pairs imparts strong generalization to CLIP model. Since training on a similar scale for videos is infeasible, recent approaches focus on the effective transfer of image-based CLIP to the video domain. In this pursuit, new parametric modules are added to learn temporal information and inter-frame relationships which require meticulous design efforts. Furthermore, when the resulting models are learned on videos, they tend to overfit on the given task distribution and lack in generalization aspect. This begs the following question: How to effectively transfer image-level CLIP representations to videos? In this work, we show that a simple Video Fine-tuned CLIP (ViFi-CLIP) baseline is generally sufficient to bridge the domain gap from images to videos. Our qualitative analysis illustrates that the framelevel processing from CLIP image-encoder followed by feature pooling and similarity matching with corresponding text embeddings helps in implicitly modeling the temporal cues within ViFi-CLIP. Such fine-tuning helps the model to focus on scene dynamics, moving objects and inter-object relationships. For low-data regimes where full fine-tuning is not viable, we propose a bridge and prompt approach that first uses fine-tuning to bridge the domain gap and then learns prompts on language and vision side to adapt CLIP representations. We extensively evaluate this simple yet strong baseline on zero-shot, base-to-novel generalization, few-shot and fully supervised settings across five video benchmarks. Our code and pre-trained models are available at https://github.com/muzairkhattak/ViFi-CLIP.

The synthesis of high-resolution remote sensing images based on text descriptions has great potential in many practical application scenarios. Although deep neural networks have achieved great success in many important remote sensing tasks, generating realistic remote sensing images from text descriptions is still very difficult. To address this challenge, we propose a novel text-to-image modern Hopfield network (Txt2Img-MHN). The main idea of Txt2Img-MHN is to conduct hierarchical prototype learning on both text and image embeddings with modern Hopfield layers. Instead of directly learning concrete but highly diverse text-image joint feature representations for different semantics, Txt2Img-MHN aims to learn the most representative prototypes from text-image embeddings, achieving a coarse-to-fine learning strategy. These learned prototypes can then be utilized to represent more complex semantics in the text-to-image generation task. To better evaluate the realism and semantic consistency of the generated images, we further conduct zero-shot classification on real remote sensing data using the classification model trained on synthesized images. Despite its simplicity, we find that the overall accuracy in the zero-shot classification may serve as a good metric to evaluate the ability to generate an image from text. Extensive experiments on the benchmark remote sensing text-image dataset demonstrate that the proposed Txt2Img-MHN can generate more realistic remote sensing images than existing methods. Code and pre-trained models are available online ( https://github.com/YonghaoXu/Txt2Img-MHN ).

Transformer models have achieved outstanding results on a variety of language tasks, such as text classification, ma- chine translation, and question answering. This success in the field of Natural Language Processing (NLP) has sparked interest in the computer vision community to apply these models to vision and multi-modal learning tasks. However, visual data has a unique structure, requiring the need to rethink network designs and training methods. As a result, Transformer models and their variations have been suc- cessfully used for image recognition, object detection, seg- mentation, image super-resolution, video understanding, image generation, text-image synthesis, and visual question answering, among other applications.

Recent advances in 3D-aware generative models (3D-aware GANs) combined with Neural Radiance Fields (NeRF) have achieved impressive results. However no prior works investigate 3D-aware GANs for 3D consistent multi-class image-to-image (3D-aware I2I) translation. Naively using 2D-I2I translation methods suffers from unrealistic shape/identity change. To perform 3D-aware multi-class I2I translation, we decouple this learning process into a multi-class 3D-aware GAN step and a 3D-aware I2I translation step. In the first step, we propose two novel techniques: a new conditional architecture and an effective training strategy. In the second step, based on the well-trained multi-class 3D-aware GAN architecture, that preserves view-consistency, we construct a 3D-aware I2I translation system. To further reduce the view-consistency problems, we propose several new techniques, including a U-net-like adaptor network design, a hierarchical representation constrain and a relative regularization loss. In extensive experiments on two datasets, quantitative and qualitative results demonstrate that we successfully perform 3D-aware I2I translation with multi-view consistency. Code is available in 3DI2I.

Burst image processing is becoming increasingly popular in recent years. However, it is a challenging task since individual burst images undergo multiple degradations and often have mutual misalignments resulting in ghosting and zipper artifacts. Existing burst restoration methods usually do not consider the mutual correlation and non-local contextual information among burst frames, which tends to limit these approaches in challenging cases. Another key challenge lies in the robust up-sampling of burst frames. The existing up-sampling methods cannot effectively utilize the advantages of single-stage and progressive up-sampling strategies with conventional and/or recent up-samplers at the same time. To address these challenges, we propose a novel Gated Multi-Resolution Transfer Network (GMTNet) to reconstruct a spatially precise high-quality image from a burst of low-quality raw images. GMT-Net consists of three modules optimized for burst processing tasks: Multi-scale Burst Feature Alignment (MBFA) for feature denoising and alignment, Transposed-Attention Feature Merging (TAFM) for multi-frame feature aggregation, and Resolution Transfer Feature Up-sampler (RTFU) to up-scale merged features and construct a high-quality output image. Detailed experimental analysis on five datasets validate our approach and sets a state-of-the-art for burst super-resolution, burst denoising, and low-light burst enhancement. Our codes and models are available at https://github.com/nanmehta/GMTNet.

Feature matching is a challenging computer vision task that involves finding correspondences between two images of a 3D scene. In this paper we consider the dense approach instead of the more common sparse paradigm, thus striving to find all correspondences. Perhaps counter-intuitively, dense methods have previously shown inferior performance to their sparse and semi-sparse counterparts for estimation of two-view geometry. This changes with our novel dense method, which outperforms both dense and sparse methods on geometry estimation. The novelty is threefold: First, we propose a kernel regression global matcher. Secondly, we propose warp refinement through stacked feature maps and depthwise convolution kernels. Thirdly, we propose learning dense confidence through consistent depth and a balanced sampling approach for dense confidence maps. Through extensive experiments we confirm that our proposed dense method, Dense Kernelized Feature Matching, sets a new state-of-the-art on multiple geometry estimation benchmarks. In particular, we achieve an improvement on MegaDepth-1500 of +4.9 and +8.9 AUC@5° compared to the best previous sparse method and dense method respectively. Our code is provided at the following repository: https://github.com/Parskatt/DKM.

We propose a fast single-stage method for both image and video instance segmentation, called SipMask, that preserves the instance spatial information by performing multiple sub-region mask predictions. The main module in our method is a light-weight spatial preservation (SP) module that generates a separate set of spatial coefficients for the sub-regions within a bounding-box, enabling a better delineation of spatially adjacent instances. To better correlate mask prediction with object detection, we further propose a mask alignment weighting loss and a feature alignment scheme. In addition, we identify two issues that impede the performance of single-stage instance segmentation and introduce two modules, including a sample selection scheme and an instance refinement module, to address these two issues. Experiments are performed on both image instance segmentation dataset MS COCO and video instance segmentation dataset YouTube-VIS. On MS COCO test-dev set, our method achieves a state-of-the-art performance. In terms of real-time capabilities, it outperforms YOLACT by a gain of 3.0% (mask AP) under the similar settings, while operating at a comparable speed. On YouTube-VIS validation set, our method also achieves promising results. The source code is available at https://github.com/JialeCao001/SipMask.

Detecting breast lesion in videos is crucial for computer-aided diagnosis. Existing video-based breast lesion detection approaches typically perform temporal feature aggregation of deep backbone features based on the self-attention operation. We argue that such a strategy struggles to effectively perform deep feature aggregation and ignores the useful local information. To tackle these issues, we propose a spatial-temporal deformable attention based framework, named STNet. Our STNet introduces a spatial-temporal deformable attention module to perform local spatial-temporal feature fusion. The spatial-temporal deformable attention module enables deep feature aggregation in each stage of both encoder and decoder. To further accelerate the detection speed, we introduce an encoder feature shuffle strategy for multi-frame prediction during inference. In our encoder feature shuffle strategy, we share the backbone and encoder features, and shuffle encoder features for decoder to generate the predictions of multiple frames. The experiments on the public breast lesion ultrasound video dataset show that our STNet obtains a state-of-the-art detection performance, while operating twice as fast inference speed. The code and model are available at https://github.com/AlfredQin/STNet.

In this work, we propose a few-shot colorectal tissue image generation method for addressing the scarcity of histopathological training data for rare cancer tissues. Our few-shot generation method, named XM-GAN, takes one base and a pair of reference tissue images as input and generates high-quality yet diverse images. Within our XM-GAN, a novel controllable fusion block densely aggregates local regions of reference images based on their similarity to those in the base image, resulting in locally consistent features. To the best of our knowledge, we are the first to investigate few-shot generation in colorectal tissue images. We evaluate our few-shot colorectral tissue image generation by performing extensive qualitative, quantitative and subject specialist (pathologist) based evaluations. Specifically, in specialist-based evaluation, pathologists could differentiate between our XM-GAN generated tissue images and real images only 55% time. Moreover, we utilize these generated images as data augmentation to address the few-shot tissue image classification task, achieving a gain of 4.4% in terms ofmean accuracy over the vanilla few-shot classifier. Code: https://github.com/VIROBO-15/XM-GAN.

Multi-label zero-shot learning strives to classify images into multiple unseen categories for which no data is available during training. The test samples can additionally contain seen categories in the generalized variant. Existing approaches rely on learning either shared or label-specific attention from the seen classes. Nevertheless, computing reliable attention maps for unseen classes during inference in a multi-label setting is still a challenge. In contrast, state-of-the-art single-label generative adversarial network (GAN) based approaches learn to directly synthesize the class-specific visual features from the corresponding class attribute embeddings. However, synthesizing multi-label features from GANs is still unexplored in the context of zero-shot setting. When multiple objects occur jointly in a single image, a critical question is how to effectively fuse multi-class information. In this work, we introduce different fusion approaches at the attribute-level, feature-level and cross-level (across attribute and feature-levels) for synthesizing multi-label features from their corresponding multi-label class embeddings. To the best of our knowledge, our work is the first to tackle the problem of multi-label feature synthesis in the (generalized) zero-shot setting. Our cross-level fusion-based generative approach outperforms the state-of-the-art on three zero-shot benchmarks: NUS-WIDE, Open Images and MS COCO. Furthermore, we show the generalization capabilities of our fusion approach in the zero-shot detection task on MS COCO, achieving favorable performance against existing methods.

The Visual Object Tracking Segmentation VOTS2023 challenge is the eleventh annual tracker benchmarking activity of the VOT initiative. This challenge is the first to merge short-term and long-term as well as single-target and multiple-target tracking with segmentation masks as the only target location specification. A new dataset was created; the ground truth has been withheld to prevent overfitting. New performance measures and evaluation protocols have been created along with a new toolkit and an evaluation server. Results of the presented 47 trackers indicate that modern tracking frameworks are well-suited to deal with convergence of short-term and long-term tracking and that multiple and single target tracking can be considered a single problem. A leaderboard, with participating trackers details, the source code, the datasets, and the evaluation kit are publicly available at the challenge website(1).

Given the often enormous effort required to train GANs, both computationally as well as in dataset collection, the re-use of pretrained GANs largely increases the potential impact of generative models. Therefore, we propose a novel knowledge transfer method for generative models based on mining the knowledge that is most beneficial to a specific target domain, either from a single or multiple pretrained GANs. This is done using a miner network that identifies which part of the generative distribution of each pretrained GAN outputs samples closest to the target domain. Mining effectively steers GAN sampling towards suitable regions of the latent space, which facilitates the posterior finetuning and avoids pathologies of other methods, such as mode collapse and lack of flexibility. Furthermore, to prevent overfitting on small target domains, we introduce sparse subnetwork selection, that restricts the set of trainable neurons to those that are relevant for the target dataset. We perform comprehensive experiments on several challenging datasets using various GAN architectures (BigGAN, Progressive GAN, and StyleGAN) and show that the proposed method, called MineGAN, effectively transfers knowledge to domains with few target images, outperforming existing methods. In addition, MineGAN can successfully transfer knowledge from multiple pretrained GANs. MineGAN.

Prompt learning has emerged as an efficient alternative for fine-tuning foundational models, such as CLIP, for various downstream tasks. Conventionally trained using the task-specific objective, i.e., cross-entropy loss, prompts tend to overfit downstream data distributions and find it challenging to capture task-agnostic general features from the frozen CLIP. This leads to the loss of the model's original generalization capability. To address this issue, our work introduces a self-regularization framework for prompting called PromptSRC (Prompting with Self-regulating Constraints). PromptSRC guides the prompts to optimize for both task-specific and task-agnostic general representations using a three-pronged approach by: (a) regulating prompted representations via mutual agreement maximization with the frozen model, (b) regulating with selfensemble of prompts over the training trajectory to encode their complementary strengths, and (c) regulating with textual diversity to mitigate sample diversity imbalance with the visual branch. To the best of our knowledge, this is the first regularization framework for prompt learning that avoids overfitting by jointly attending to pre-trained model features, the training trajectory during prompting, and the textual diversity. PromptSRC explicitly steers the prompts to learn a representation space that maximizes performance on downstream tasks without compromising CLIP generalization. We perform extensive experiments on 4 benchmarks where PromptSRC overall performs favorably well compared to the existing methods. Our code and pre-trained models are publicly available at: https://github.com/muzairkhattak/PromptSRC.

We present a method to efficiently generate 3D-aware high-resolution images that are view-consistent across multiple target views. The proposed multiplane neural radiance model, named GMNR, consists of a novel a-guided view-dependent representation (a-VdR) module for learning view-dependent information. The a-VdR module, faciliated by an a-guided pixel sampling technique, computes the view-dependent representation efficiently by learning viewing direction and position coefficients. Moreover, we propose a view-consistency loss to enforce photometric similarity across multiple views. The GMNR model can generate 3D-aware high-resolution images that are view-consistent across multiple camera poses, while maintaining the computational efficiency in terms of both training and inference time. Experiments on three datasets demonstrate the effectiveness of the proposed modules, leading to favorable results in terms of both generation quality and inference time, compared to existing approaches. Our GMNR model generates 3D-aware images of 1024 x 1024 pixels with 17.6 FPS on a single V100. Code : https: //github.com/VIROBO-15/GMNR

For most applications, 2D keypoint detection works well and offers a simple and fast tool to analyse human movements. However, there remain many situations where even the best state-of-the-art algorithms reach their limits and fail to detect human keypoints correctly. Such situations may occur especially when individual body parts are occluded, twisted, or when the whole person is flipped. Especially when analysing injuries in alpine ski racing, such twisted and rotated body positions occur frequently. To improve the detection of keypoints for this application, we developed a novel method that refines keypoint estimates by rotating the input videos. We select the best rotation for every frame with a graph-based global solver. Thereby, we improve keypoint detection of an arbitrary pose estimation algorithm, in particular for 'hard' keypoints. In the current proof-of-concept study, we show that our approach outperforms standard keypoint detection results in all categories and in all metrics, in injury-related out-of-balance and fall situations by a large margin as well as previous methods, in performance and robustness. The Injury Ski II dataset was made publicly available, aiming to facilitate the investigation of sports accidents based on computer vision in the future.

Connected and autonomous vehicles (CAVs) are an emerging trend in the transport sector and their impact on transportation, the economy, society and the environment will be tremendous. Much like the automobile shaped the way humans travelled, lived and worked during the 20th century, CAVs have yet again the potential to affect and reform all of these areas. Besides the imminent technological challenges on the robotic aspect of making CAVs become a market-ready reality, a plethora of ethical, social and legal questions will have to be addressed along the line. Knowledge of and interaction with the surrounding infrastructure and other actors in the system will be essential for CAVs in order to pave the way for progressive solutions to urgent sustainability and mobility issues in transportation.

Road networks, i.e. the networks of roads and intersections, are the core infrastructure on which CAVs will operate. Thus, having detailed knowledge about them is key for CAVs in order to take the right decisions on both short-term actions that will affect individual traffic users in immediate situations and long-term actions that will affect entire transportation systems in the long run. Machine learning is nowadays a popular choice to extract and conglomerate knowledge from large amounts of data – and large amounts of data can be obtained about road networks. However, classical machine learning models are incapable of harnessing the graph-structured nature of road networks sufficiently.

Graph neural networks (GNNs) are machine learning models of growing popularity that can explicitly leverage the complex topological structure of node dependencies in graphs, such as the ones observed in road networks. Road networks are sparse graphs that reside in a euclidean space, and therefore different to typical graphs studied in the literature. Also, crowd-sourced road network graphs often have incomplete attributes and are generally lacking the fine-grained level of detail in their encoded information that would be required for CAVs. Identifying the best representation of road network graphs and complementing their lacking detail with auxiliary data is therefore an important research direction.

This thesis, therefore, addresses data-driven classification in road networks from two directions: A) the general approach of learning on spatial graphs of road network with GNNs, and B) complementing road network graphs with auxiliary data. Specifically, this thesis and the included papers address the exemplary task of road classifications and make the following contributions to the field:

Paper A analyses how GNNs can be applied to road networks and how the networks are best represented. Different aggregator functions are compared on final classification performances. A novel aggregator and a neighbourhood sampling method are introduced, and the line graph transformation is identified as a suitable representation of road network graphs for GNNs.

Paper B complements the road network graphs with mobility data from millions of GPS trajectories and introduces an equitemporal node spacing to create road segments of equal travel time. It further introduces remote sensing vision data as a potent complement to overcome shortcom-ings of the graph-based representation for road networks. Simple hand-crafted low-level vision features are used in this work. However, both the equitemporal node spacing and the simple vision features clearly exhibit improved classification performances.

Finally, Paper C consolidates the complement of remote sensing data to the road network graphs. Through a general visual feature encoding of state-of-the-art pretrained vision back-bones that are carefully fine-tuned to the remote sensing domain, a further performance boost on the road classification task is achieved.

This article proposes an architecture, which allows the prediction of intention by internally simulating perceptual states represented by action pattern vectors. To this end, associative self-organising neural networks (A-SOM) is utilised to build a hierarchical cognitive architecture for recognition and simulation of the skeleton based human actions. The abilities of the proposed architecture in recognising and predicting actions is evaluated in experiments using three different datasets of 3D actions. Based on the experiments of this article, applying internally simulated perceptual states represented by action pattern vectors improves the performance of the recognition task in all experiments. Furthermore, internal simulation of perception addresses the problem of having limited access to the sensory input, and also the future prediction of the consecutive perceptual sequences. The performance of the system is compared and discussed with similar architecture using self-organizing neural networks (SOM).

Adversarial training (AT) is an effective approach to making deep neural networks robust against adversarial attacks. Recently, different AT defenses are proposed that not only maintain a high clean accuracy but also show significant robustness against popular and well-studied adversarial attacks, such as projected gradient descent (PGD). High adversarial robustness can also arise if an attack fails to find adversarial gradient directions, a phenomenon known as "gradient masking." In this work, we analyze the effect of label smoothing on AT as one of the potential causes of gradient masking. We then develop a guided mechanism to avoid local minima during attack optimization, leading to a novel attack dubbed guided projected gradient attack (G-PGA). Our attack approach is based on a "match and deceive" loss that finds optimal adversarial directions through guidance from a surrogate model. Our modified attack does not require random restarts a large number of attack iterations or a search for optimal step size. Furthermore, our proposed G-PGA is generic, thus it can be combined with an ensemble attack strategy as we demonstrate in the case of auto-attack, leading to efficiency and convergence speed improvements. More than an effective attack, G-PGA can be used as a diagnostic tool to reveal elusive robustness due to gradient masking in adversarial defenses.

In recent years, Siamese network based trackers have significantly advanced the state-of-the-art in real-time tracking. Despite their success, Siamese trackers tend to suffer from high memory costs, which restrict their applicability to mobile devices with tight memory budgets. To address this issue, we propose a distilled Siamese tracking framework to learn small, fast and accurate trackers (students), which capture critical knowledge from large Siamese trackers (teachers) by a teacher-students knowledge distillation model. This model is intuitively inspired by the one teacher versus multiple students learning method typically employed in schools. In particular, our model contains a single teacher-student distillation module and a student-student knowledge sharing mechanism. The former is designed using a tracking-specific distillation strategy to transfer knowledge from a teacher to students. The latter is utilized for mutual learning between students to enable in-depth knowledge understanding. Extensive empirical evaluations on several popular Siamese trackers demonstrate the generality and effectiveness of our framework. Moreover, the results on five tracking benchmarks show that the proposed distilled trackers achieve compression rates of up to 18x and frame-rates of 265 FPS, while obtaining comparable tracking accuracy compared to base models.

Modern digital cameras generally count on image signal processing (ISP) pipelines for producing naturalistic RGB images. Nevertheless, in comparison to DSLR cameras, low-quality images are generally output from portable mobile devices due to their physical limitations. The synthesized low-quality images usually have multiple degradations - low-resolution owing to small camera sensors, mosaic patterns on account of camera filter array and subpixel shifts due to camera motion. Such degradation usually restrain the performance of single image super-resolution methodologies for retrieving high-resolution (HR) image from a single low-resolution (LR) image. Burst image super-resolution aims at restoring a photo-realistic HR image by capturing the abundant information from multiple LR images. Lately, the soaring popularity of burst photography has made multi-frame processing an attractive solution for overcoming the limitations of single image processing. In our work, we thus aim to propose a generic architecture, adaptive feature consolidation network (AFCNet) for multi-frame processing. To alleviate the challenge of effectively modelling the long-range dependency problem, that multi-frame approaches struggle to solve, we utilize encoder-decoder based transformer backbone which learns multi-scale local-global representations. We propose feature alignment module to align LR burst frame features. Further, the aligned features are fused and reconstructed by abridged pseudo-burst fusion module and adaptive group upsampling modules, respectively. Our proposed approach clearly outperforms the other existing state-of-the-art techniques on benchmark datasets. The experimental results illustrate the effectiveness and generality of our proposed framework in upgrading the visual quality of HR images.

Modern handheld devices can acquire burst image sequence in a quick succession. However, the individual acquired frames suffer from multiple degradations and are misaligned due to camera shake and object motions. The goal of Burst Image Restoration is to effectively combine complimentary cues across multiple burst frames to generate high-quality outputs. Towards this goal, we develop a novel approach by solely focusing on the effective information exchange between burst frames, such that the degradations get filtered out while the actual scene details are preserved and enhanced. Our central idea is to create a set of pseudo-burst features that combine complimentary information from all the input burst frames to seamlessly exchange information. However, the pseudo-burst cannot be successfully created unless the individual burst frames are properly aligned to discount interframe movements. Therefore, our approach initially extracts pre-processed features from each burst frame and matches them using an edge-boosting burst alignment module. The pseudo-burst features are then created and enriched using multi-scale contextual information. Our final step is to adaptively aggregate information from the pseudo-burst features to progressively increase resolution in multiple stages while merging the pseudo-burst features. In comparison to existing works that usually follow a late fusion scheme with single-stage upsampling, our approach performs favorably, delivering state-of-the-art performance on burst super-resolution, burst low-light image enhancement and burst denoising tasks. The source code and pre-trained models are available at https://github.com/akshaydudhane16/BIPNet.

Open-world object detection (OWOD) is a challenging computer vision problem, where the task is to detect a known set of object categories while simultaneously identifying unknown objects. Additionally, the model must incrementally learn new classes that become known in the next training episodes. Distinct from standard object detection, the OWOD setting poses significant challenges for generating quality candidate proposals on potentially unknown objects, separating the unknown objects from the background and detecting diverse unknown objects. Here, we introduce a novel end-to-end transformer-based framework, OW-DETR, for open-world object detection. The proposed OW-DETR comprises three dedicated components namely, attention-driven pseudo-labeling, novelty classification and objectness scoring to explicitly address the aforementioned OWOD challenges. Our OW-DETR explicitly encodes multi-scale contextual information, possesses less inductive bias, enables knowledge transfer from known classes to the unknown class and can better discriminate between unknown objects and background. Comprehensive experiments are performed on two benchmarks: MS-COCO and PASCAL VOC. The extensive ablations reveal the merits of our proposed contributions. Further, our model outperforms the recently introduced OWOD approach, ORE, with absolute gains ranging from 1.8% to 3.3% in terms of unknown recall on MS-COCO. In the case of incremental object detection, OW-DETR outperforms the state-of-theart for all settings on PASCAL VOC. Our code is available at https://github.com/akshitac8/OW-DETR.

Anomaly detection is commonly pursued as a one-class classification problem, where models can only learn from normal training samples, while being evaluated on both normal and abnormal test samples. Among the successful approaches for anomaly detection, a distinguished category of methods relies on predicting masked information (e.g. patches, future frames, etc.) and leveraging the reconstruction error with respect to the masked information as an abnormality score. Different from related methods, we propose to integrate the reconstruction-based functionality into a novel self-supervised predictive architectural building block. The proposed self-supervised block is generic and can easily be incorporated into various state-of-the-art anomaly detection methods. Our block starts with a convolutional layer with dilated filters, where the center area of the receptive field is masked. The resulting activation maps are passed through a channel attention module. Our block is equipped with a loss that minimizes the reconstruction error with respect to the masked area in the receptive field. We demonstrate the generality of our block by integrating it into several state-of-the-art frameworks for anomaly detection on image and video, providing empirical evidence that shows considerable performance improvements on MVTec AD, Avenue, and ShanghaiTech. We release our code as open source at: https://github.com/ ristea/sspcab.

In this paper, we propose self:supervised training for video transformers using unlabeled video data. From a given video, we create local and global spatiotemporal views with varying spatial sizes and frame rates. Our self-supervised objective seeks to match the features of these different views representing the same video, to be invariant to spatiotemporal variations in actions. To the best of our knowledge, the proposed approach is the first to alleviate the dependency on negative samples or dedicated memory banks in Self-supervised Video Transformer (SVT). Further; owing to the flexibility of Transformer models, SVT supports slow-fast video processing within a single architecture using dynamically adjusted positional encoding and supports long-term relationship modeling along spatiotemporal dimensions. Our approach performs well on four action recognition benchmarks (Kinetics-400, UCF-101, HMDB-51, and SSv2) and converges faster with small batch sizes. Code is available at: https://git.io/J1juJ

Detecting abnormal events in video is commonly framed as a one-class classification task, where training videos contain only normal events, while test videos encompass both normal and abnormal events. In this scenario, anomaly detection is an open-set problem. However, some studies assimilate anomaly detection to action recognition. This is a closed-set scenario that fails to test the capability of systems at detecting new anomaly types. To this end, we propose UBnormal, a new supervised open-set benchmark composed of multiple virtual scenes for video anomaly detection. Unlike existing data sets, we introduce abnormal events annotated at the pixel level at training time, for the first time enabling the use of fully-supervised learning methods for abnormal event detection. To preserve the typical open-set formulation, we make sure to include dis-joint sets of anomaly types in our training and test collections of videos. To our knowledge, UBnormal is the first video anomaly detection benchmark to allow a fair head-to-head comparison between one-class open-set models and supervised closed-set models, as shown in our experiments. Moreover, we provide empirical evidence showing that UB-normal can enhance the performance of a state-of-the-art anomaly detection framework on two prominent data sets, Avenue and ShanghaiTech. Our benchmark is freely available at https://github.com/lilygeorgescu/UBnormal.

Road networks are the core infrastructure for connected and autonomous vehicles, but creating meaningful representations for machine learning applications is a challenging task. In this work, we propose to integrate remote sensing vision data into road network data for improved embeddings with graph neural networks. We present a segmentation of road edges based on spatio-temporal road and traffic characteristics, which allows enriching the attribute set of road networks with visual features of satellite imagery and digital surface models. We show that both, the segmentation and the integration of vision data can increase performance on a road type classification task, and we achieve state-of-the-art performance on the OSM+DiDi Chuxing dataset on Chengdu, China.

Pedestrian detection is an important but challenging problem in computer vision, especially in human-centric tasks. Over the past decade, significant improvement has been witnessed with the help of handcrafted features and deep features. Here we present a comprehensive survey on recent advances in pedestrian detection. First, we provide a detailed review of single-spectral pedestrian detection that includes handcrafted features based methods and deep features based approaches. For handcrafted features based methods, we present an extensive review of approaches and find that handcrafted features with large freedom degrees in shape and space have better performance. In the case of deep features based approaches, we split them into pure CNN based methods and those employing both handcrafted and CNN based features. We give the statistical analysis and tendency of these methods, where feature enhanced, part-aware, and post-processing methods have attracted main attention. In addition to single-spectral pedestrian detection, we also review multi-spectral pedestrian detection, which provides more robust features for illumination variance. Furthermore, we introduce some related datasets and evaluation metrics, and a deep experimental analysis. We conclude this survey by emphasizing open problems that need to be addressed and highlighting various future directions. Researchers can track an up-to-date list at https://github.com/JialeCao001/PedSurvey.

State-of-the-art transformer-based video instance segmentation (VIS) approaches typically utilize either single-scale spatio-temporal features or per-frame multi-scale features during the attention computations. We argue that such an attention computation ignores the multiscale spatio-temporal feature relationships that are crucial to tackle target appearance deformations in videos. To address this issue, we propose a transformer-based VIS framework, named MS-STS VIS, that comprises a novel multi-scale spatio-temporal split (MS-STS) attention module in the encoder. The proposed MS-STS module effectively captures spatio-temporal feature relationships at multiple scales across frames in a video. We further introduce an attention block in the decoder to enhance the temporal consistency of the detected instances in different frames of a video. Moreover, an auxiliary discriminator is introduced during training to ensure better foreground-background separability within the multiscale spatio-temporal feature space. We conduct extensive experiments on two benchmarks: Youtube-VIS (2019 and 2021). Our MS-STS VIS achieves state-of-the-art performance on both benchmarks. When using the ResNet50 backbone, our MS-STS achieves a mask AP of 50.1%, outperforming the best reported results in literature by 2.7% and by 4.8% at higher overlap threshold of AP75, while being comparable in model size and speed on Youtube-VIS 2019 val. set. When using the Swin Transformer backbone, MS-STS VIS achieves mask AP of 61.0% on Youtube-VIS 2019 val. set.

Following the successful application of vision transformers in multiple computer vision tasks, these models have drawn the attention of the signal processing community. This is because signals are often represented as spectrograms (e.g. through Discrete Fourier Transform) which can be directly provided as input to vision transformers. However, naively applying transformers to spectrograms is suboptimal. Since the axes represent distinct dimensions, i.e. frequency and time, we argue that a better approach is to separate the attention dedicated to each axis. To this end, we propose the Separable Transformer (SepTr), an architecture that employs two transformer blocks in a sequential manner, the first attending to tokens within the same time interval, and the second attending to tokens within the same frequency bin. We conduct experiments on three benchmark data sets, showing that our separable architecture outperforms conventional vision transformers and other state-of-the-art methods. Unlike standard transformers, SepTr linearly scales the number of trainable parameters with the input size, thus having a lower memory footprint. Our code is available as open source at https://github.com/ristea/septr.

Automatically identifying harmful content in video is an important task with a wide range of applications. However, there is a lack of professionally labeled open datasets available. In this work VidHarm, an open dataset of 3589 video clips from film trailers annotated by professionals, is presented. An analysis of the dataset is performed, revealing among other things the relation between clip and trailer level annotations. Audiovisual models are trained on the dataset and an in-depth study of modeling choices conducted. The results show that performance is greatly improved by combining the visual and audio modality, pre-training on large-scale video recognition datasets, and class balanced sampling. Lastly, biases of the trained models are investigated using discrimination probing. VidHarm is openly available, and further details are available at the webpage https://vidharm.github.io/

Face alignment is the process of determining a face shape given its location and size in an image. It is used as a basis for other facial analysis tasks and for human-machine interaction and augmented reality applications. It is a challenging problem due to the extremely high variability in facial appearance affected by many external (illumination, occlusion, head pose) and internal factors (race, facial expression). However, advances in deep learning combined with domain-related knowledge from previous research recently demonstrated impressive results nearly saturating the unconstrained benchmark data sets. The focus is shifting towards reducing the computational burden of the face alignment models since real-time performance is required for such a highly dynamic task. Furthermore, many applications target devices on the edge with limited computational power which puts even greater emphasis on computational efficiency. We present the latest development in regression-based approaches that have led towards nearly solving the face alignment problem in an unconstrained scenario. Various regression architectures are systematically explored and recent training techniques discussed in the context of face alignment. Finally, a benchmark comparison of the most successful methods is presented, taking into account execution time as well, to provide a comprehensive overview of this dynamic research field. (C) 2020 Elsevier B.V. All rights reserved.

In the last decade, developments in hardware, sensors and software have made it possible to create increasingly autonomous systems. These systems can be as simple as limited driver assistance software lane-following in cars, or limited collision warning systems for otherwise manually piloted drones. On the other end of the spectrum exist fully autonomous cars, boats or helicopters. With increasing abilities to function autonomously, the demands to operate with minimal human supervision in unstructured environments increase accordingly.

Common to most, if not all, autonomous systems is that they require an accurate model of the surrounding world. While there is currently a large number of possible sensors useful to create such models available, cameras are one of the most versatile. From a sensing perspective cameras have several advantages over other sensors in that they require no external infrastructure, are relatively cheap and can be used to extract such information as the relative positions of other objects, their movements over time, create accurate maps and locate the autonomous system within these maps.

Using cameras to produce a model of the surroundings require solving a number of technical problems. Often these problems have a basis in recognizing that an object or region of interest is the same over time or in novel viewpoints. In visual tracking this type of recognition is required to follow an object of interest through a sequence of images. In geometric problems it is often a requirement to recognize corresponding image regions in order to perform 3D reconstruction or localization. 

The first set of contributions in this thesis is related to the improvement of a class of on-line learned visual object trackers based on discriminative correlation filters. In visual tracking estimation of the objects size is important for reliable tracking, the first contribution in this part of the thesis investigates this problem. The performance of discriminative correlation filters is highly dependent on what feature representation is used by the filter. The second tracking contribution investigates the performance impact of different features derived from a deep neural network.

A second set of contributions relate to the evaluation of visual object trackers. The first of these are the visual object tracking challenge. This challenge is a yearly comparison of state-of-the art visual tracking algorithms. A second contribution is an investigation into the possible issues when using bounding-box representations for ground-truth data.

In real world settings tracking typically occur over longer time sequences than is common in benchmarking datasets. In such settings it is common that the model updates of many tracking algorithms cause the tracker to fail silently. For this reason it is important to have an estimate of the trackers performance even in cases when no ground-truth annotations exist. The first of the final three contributions investigates this problem in a robotics setting, by fusing information from a pre-trained object detector in a state-estimation framework. An additional contribution describes how to dynamically re-weight the data used for the appearance model of a tracker. A final contribution investigates how to obtain an estimate of how certain detections are in a setting where geometrical limitations can be imposed on the search region. The proposed solution learns to accurately predict stereo disparities along with accurate assessments of each predictions certainty.

In computer vision, the aim is to model and extract high-level information from visual sensor measurements such as images, videos and 3D points. Since visual data is often high-dimensional, noisy and irregular, achieving robust data modeling is challenging. This thesis presents works that address challenges within a number of different computer vision problems. 

First, the thesis addresses the problem of phase unwrapping for multi-frequency amplitude modulated time-of-flight (ToF) ranging. ToF is used in depth cameras, which have many applications in 3D reconstruction and gesture recognition. While amplitude modulation in time-of-flight ranging can provide accurate measurements for the depth, it also causes depth ambiguities. This thesis presents a method to resolve the ambiguities by estimating the likelihoods of different hypotheses for the depth values. This is achieved by performing kernel density estimation over the hypotheses in a spatial neighborhood of each pixel in the depth image. The depth hypothesis with the highest estimated likelihood can then be selected as the output depth. This approach yields improvements in the quality of the depth images and extends the effective range in both indoor and outdoor environments. 

Next, point set registration is investigated, which is the problem of aligning point sets from overlapping depth images or 3D models. Robust registration is fundamental to many vision tasks, such as multi-view 3D reconstruction and object pose estimation for robotics. The thesis presents a method for handling density variations in the measured point sets. This is achieved by modeling a latent distribution representing the underlying structure of the scene. Both the model of the scene and the registration parameters are inferred in an Expectation-Maximization based framework. Secondly, the thesis introduces a method for integrating features from deep neural networks into the registration model. It is shown that the deep features improve registration performance in terms of accuracy and robustness. Additionally, improved feature representations are generated by training the deep neural network end-to-end by minimizing registration errors produced by our registration model. 

Further, an approach for 3D point set segmentation is presented. As scene models are often represented using 3D point measurements, segmentation of these is important for general scene understanding. Learning models for segmentation requires a significant amount of annotated data, which is expensive and time-consuming to acquire. The approach presented in the thesis circumvents this by projecting the points into virtual camera views and render 2D images. The method can then exploit accurate convolutional neural networks for image segmentation and map the segmentation predictions back to the 3D points. This also allows for transferring learning using available annotated image data, thereby reducing the need for 3D annotations. 

Finally, the thesis explores the problem of video object segmentation (VOS), where the task is to track and segment target objects in each frame of a video sequence. Accurate VOS requires a robust model of the target that can adapt to different scenarios and objects. This needs to be achieved using only a single labeled reference frame as training data for each video sequence. To address the challenges in VOS, the thesis introduces a parametric target model, optimized to predict a target label derived from the mask annotation. The target model is integrated into a deep neural network, where its predictions guide a decoder module to produce target segmentation masks. The deep network is trained on labeled video data to output accurate segmentation masks for each frame. Further, it is shown that by training the entire network model in an end-to-end manner, it can learn a representation of the target that provides increased segmentation accuracy. 

Humans have a natural instinct to identify unknown object instances in their environments. The intrinsic curiosityabout these unknown instances aids in learning about them,when the corresponding knowledge is eventually available.This motivates us to propose a novel computer vision problem called: ‘Open World Object Detection’, where a modelis tasked to: 1) identify objects that have not been introduced to it as ‘unknown’, without explicit supervision to doso, and 2) incrementally learn these identified unknown categories without forgetting previously learned classes, whenthe corresponding labels are progressively received. Weformulate the problem, introduce a strong evaluation protocol and provide a novel solution, which we call ORE:Open World Object Detector, based on contrastive clustering and energy based unknown identification. Our experimental evaluation and ablation studies analyse the efficacyof ORE in achieving Open World objectives. As an interesting by-product, we find that identifying and characterisingunknown instances helps to reduce confusion in an incremental object detection setting, where we achieve state-ofthe-art performance, with no extra methodological effort.We hope that our work will attract further research into thisnewly identified, yet crucial research direction.

We propose a novel transformer-based styled handwritten text image generation approach, HWT, that strives to learn both style-content entanglement as well as global and local writing style patterns. The proposed HWT captures the long and short range relationships within the style examples through a self-attention mechanism, thereby encoding both global and local style patterns. Further, the proposed transformer-based HWT comprises an encoder-decoder attention that enables style-content entanglement by gathering the style representation of each query character. To the best of our knowledge, we are the first to introduce a transformer-based generative network for styled handwritten text generation. Our proposed HWT generates realistic styled handwritten text images and significantly outperforms the state-of-the-art demonstrated through extensive qualitative, quantitative and human-based evaluations. The proposed HWT can handle arbitrary length of text and any desired writing style in a few-shot setting. Further, our HWT generalizes well to the challenging scenario where both words and writing style are unseen during training, generating realistic styled handwritten text images.

This work proposes a weakly-supervised temporal action localization framework, called D2-Net, which strives to temporally localize actions using video-level supervision. Our main contribution is the introduction of a novel loss formulation, which jointly enhances the discriminability of latent embeddings and robustness of the output temporal class activations with respect to foreground-background noise caused by weak supervision. The proposed formulation comprises a discriminative and a denoising loss term for enhancing temporal action localization. The discriminative term incorporates a classification loss and utilizes a top-down attention mechanism to enhance the separability of latent foreground-background embeddings. The denoising loss term explicitly addresses the foreground-background noise in class activations by simultaneously maximizing intra-video and inter-video mutual information using a bottom-up attention mechanism. As a result, activations in the foreground regions are emphasized whereas those in the background regions are suppressed, thereby leading to more robust predictions. Comprehensive experiments are performed on multiple benchmarks, including THUMOS14 and ActivityNet1.2. Our D2-Net performs favorably in comparison to the existing methods on all datasets, achieving gains as high as 2.3% in terms of mAP at IoU=0.5 on THUMOS14

The choice of the material base to which the material decomposition is performed in dual-energy computed tomography may affect the quality of reconstructed images. The aim of this work is to investigate how the commonly used bases (water, bone), (water, iodine) and (photoelectric effect, Compton scattering) affect the reconstructed linear attenuation coefficient in the case of the Alvarez–Macovski method. The performance of this method is also compared with the performance of the Dual-energy Iterative Reconstruction Algorithm (DIRA). In both cases, the study is performed using simulations. The results show that the Alvarez–Macovski method produced artefacts when iodine was present in the phantom together with human tissues since this method can only work with one doublet. It was shown that these artefacts could be avoided with DIRA using the (water, bone) doublet for tissues and the (water, iodine) doublet for the iodine solution.

Safe robot navigation in a dynamic environment, requires the trajectories of each independently moving object (IMO). We present the novel and effective system Sequential Hierarchical Ransac Estimation (Shire) designed for this purpose. The system uses a stereo camera stream to find the objects and trajectories in real time. Shire detects moving objects using geometric consistency and finds their trajectories using bundle adjustment. Relying on geometric consistency allows the system to handle objects regardless of semantic class, unlike approaches based on semantic segmentation. Most Visual Odometry (VO) systems are inherently limited to single motion by the choice of tracker. This limitation allows for efficient and robust ego-motion estimation in real time, but preclude tracking the multiple motions sought. Shire instead uses a generic tracker and achieves accurate VO and IMO estimates using track analysis. This removes the restriction to a single motion while retaining the real-time performance required for live navigation. We evaluate the system by bounding box intersection over union and ID persistence on a public dataset, collected from an autonomous test vehicle driving in real traffic. We also show the velocities of estimated IMOs. We investigate variations of the system that provide trade offs between accuracy, performance and limitations.

The Visual Object Tracking challenge VOT2021 is the ninth annual tracker benchmarking activity organized by the VOT initiative. Results of 71 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in recent years. The VOT2021 challenge was composed of four sub-challenges focusing on different tracking domains: (i) VOT-ST2021 challenge focused on short-term tracking in RGB, (ii) VOT-RT2021 challenge focused on "real-time" short-term tracking in RGB, (iii) VOT-LT2021 focused on long-term tracking, namely coping with target disappearance and reappearance and (iv) VOT-RGBD2021 challenge focused on long-term tracking in RGB and depth imagery. The VOT-ST2021 dataset was refreshed, while VOT-RGBD2021 introduces a training dataset and sequestered dataset for winner identification. The source code for most of the trackers, the datasets, the evaluation kit and the results along with the source code for most trackers are publicly available at the challenge website(1).

Video instance segmentation is one of the core problems in computer vision. Formulating a purely learning-based method, which models the generic track management required to solve the video instance segmentation task, is a highly challenging problem. In this work, we propose a novel learning framework where the entire video instance segmentation problem is modeled jointly. To this end, we design a graph neural network that in each frame jointly processes all detections and a memory of previously seen tracks. Past information is considered and processed via a recurrent connection. We demonstrate the effectiveness of the proposed approach in comprehensive experiments. Our approach, operating at over 25 FPS, outperforms previous video real-time methods. We further conduct detailed ablative experiments that validate the different aspects of our approach.

Dual-energy computed tomography (CT) can be used in radiotherapy treatment planning for the calculation of absorbed dose distributions. The aim of this work is to evaluate whether there is room for improvement in the accuracy of the Monoenergetic Plus algorithm by Siemens Healthineers. A Siemens SOMATOM Force scanner was used to scan a cylindrical polymethyl methacrylate phantom with four rod-inserts made of different materials. Images were reconstructed using ADMIRE and processed with Monoenergetic Plus. The resulting CT numbers were compared with tabulated values and values simulated by the proof-of-a-concept algorithm DIRA developed by the authors. Both the Monoenergetic Plus and DIRA algorithms performed well; the accuracy of attenuation coefficients was better than about ±1% at the energy of 70 keV. Compared with DIRA, the worse performance of Monoenergetic Plus was caused by its (i) two-material decomposition to iodine and water and (ii) imperfect suppression of the beam hardening artifact in ADMIRE.

This work proposes a weakly-supervised temporal action localization framework, called D2-Net, which strives to temporally localize actions using video-level supervision. Our main contribution is the introduction of a novel loss formulation, which jointly enhances the discriminability of latent embeddings and robustness of the output temporal class activations with respect to foreground-background noise caused by weak supervision. The proposed formulation comprises a discriminative and a denoising loss term for enhancing temporal action localization. The discriminative term incorporates a classification loss and utilizes a top-down attention mechanism to enhance the separability of latent foreground-background embeddings. The denoising loss term explicitly addresses the foreground-background noise in class activations by simultaneously maximizing intra-video and inter-video mutual information using a bottom-up attention mechanism. As a result, activations in the foreground regions are emphasized whereas those in the background regions are suppressed, thereby leading to more robust predictions. Comprehensive experiments are performed on multiple benchmarks, including THUMOS14 and ActivityNet1.2. Our D2-Net performs favorably in comparison to the existing methods on all datasets, achieving gains as high as 2.3% in terms of mAP at IoU=0.5 on THUMOS14. Source code is available at https://github.com/naraysa/D2-Net.

In this paper we consider a collection of relative pose problems which arise naturally in applications for visual indoor navigation using unmanned aerial vehicles (UAVs). We focus on cases where additional information from an onboard IMU is available and thus provides a partial extrinsic calibration through the gravitational vector. The solvers are designed for a partially calibrated camera, for a variety of realistic indoor scenarios, which makes it possible to navigate using images of the ground floor. Current state-of-the-art solvers use more general assumptions, such as using arbitrary planar structures; however, these solvers do not yield adequate reconstructions for real scenes, nor do they perform fast enough to be incorporated in real-time systems. We show that the proposed solvers enjoy better numerical stability, are faster, and require fewer point correspondences, compared to state-of-the-art approaches. These properties are vital components for robust navigation in real-time systems, and we demonstrate on both synthetic and real data that our method outperforms other solvers, and yields superior motion estimation(1).

While the untargeted black-box transferability of adversarial perturbations has been extensively studied before, changing an unseen models decisions to a specific targeted class remains a challenging feat. In this paper, we propose a new generative approach for highly transferable targeted perturbations (TTP). We note that the existing methods are less suitable for this task due to their reliance on class-boundary information that changes from one model to another, thus reducing transferability. In contrast, our approach matches the perturbed image distribution with that of the target class, leading to high targeted transferability rates. To this end, we propose a new objective function that not only aligns the global distributions of source and target images, but also matches the local neighbourhood structure between the two domains. Based on the proposed objective, we train a generator function that can adaptively synthesize perturbations specific to a given input. Our generative approach is independent of the source or target domain labels, while consistently performs well against state-of-the-art methods on a wide range of attack settings. As an example, we achieve 32.63% target transferability from (an adversarially weak) VGG19(BN) to (a strong) WideResNet on ImageNet val. set, which is 4x higher than the previous best generative attack and 16x better than instance-specific iterative attack. Code is available at: https://github.com/Muzammal-Naseer/TTP.

In this paper we present a novel algorithm for onboard radial distortion correction for unmanned aerial vehicles (UAVs) equipped with an inertial measurement unit (IMU), that runs in real-time. This approach makes calibration procedures redundant, thus allowing for exchange of optics extemporaneously. By utilizing the IMU data, the cameras can be aligned with the gravity direction. This allows us to work with fewer degrees of freedom, and opens up for further intrinsic calibration. We propose a fast and robust minimal solver for simultaneously estimating the focal length, radial distortion profile and motion parameters from homographies. The proposed solver is tested on both synthetic and real data, and perform better or on par with state-of-the-art methods relying on pre-calibration procedures. Code available at: https://github.com/marcusvaltonen/HomLib. 1

In this paper, we present a state-of-the-art reinforcement learning method for autonomous driving. Our approach employs temporal difference learning in a Bayesian framework to learn vehicle control signals from sensor data. The agent has access to images from a forward facing camera, which are pre-processed to generate semantic segmentation maps. We trained our system using both ground truth and estimated semantic segmentation input. Based on our observations from a large set of experiments, we conclude that training the system on ground truth input data leads to better performance than training the system on estimated input even if estimated input is used for evaluation. The system is trained and evaluated in a realistic simulated urban environment using the CARLA simulator. The simulator also contains a benchmark that allows for comparing to other systems and methods. The required training time of the system is shown to be lower and the performance on the benchmark superior to competing approaches.

Solving geometric tasks involving point clouds by using machine learning is a challenging problem. Standard feed-forward neural networks combine linear or, if the bias parameter is included, affine layers and activation functions. Their geometric modeling is limited, which motivated the prior work introducing the multilayer hypersphere perceptron (MLHP). Its constituent part, i.e., the hypersphere neuron, is obtained by applying a conformal embedding of Euclidean space. By virtue of Clifford algebra, it can be implemented as the Cartesian dot product of inputs and weights. If the embedding is applied in a manner consistent with the dimensionality of the input space geometry, the decision surfaces of the model units become combinations of hyperspheres and make the decision-making process geometrically interpretable for humans. Our extension of the MLHP model, the multilayer geometric perceptron (MLGP), and its respective layer units, i.e., geometric neurons, are consistent with the 3D geometry and provide a geometric handle of the learned coefficients. In particular, the geometric neuron activations are isometric in 3D, which is necessary for rotation and translation equivariance. When classifying the 3D Tetris shapes, we quantitatively show that our model requires no activation function in the hidden layers other than the embedding to outperform the vanilla multilayer perceptron. In the presence of noise in the data, our model is also superior to the MLHP.

We investigate a novel deep-learning-based approach to estimate uncertainty in stereo disparity prediction networks. Current state-of-the-art methods often formulate disparity prediction as a regression problem with a single scalar output in each pixel. This can be problematic in practical applications as in many cases there might not exist a single well defined disparity, for example in cases of occlusions or at depth-boundaries. While current neural-network-based disparity estimation approaches  obtain good performance on benchmarks, the disparity prediction is treated as a black box at inference time. In this paper we show that by formulating the learning problem as a regression with a distribution target, we obtain a robust estimate of the uncertainty in each pixel, while maintaining the performance of the original method. The proposed method is evaluated both on a large-scale standard benchmark, as well on our own data. We also show that the uncertainty estimate significantly improves by maximizing the uncertainty in those pixels that have no well defined disparity during learning.

Facial expression recognition applications demand accurate and fast algorithms that can run in real time on platforms with limited computational resources. We propose an algorithm that bridges the gap between precise but slow methods and fast but less precise methods. The algorithm combines gentle boost decision trees and neural networks. The gentle boost decision trees are trained to extract highly discriminative feature vectors (local binary features) for each basic facial expression around distinct facial landmark points. These sparse binary features are concatenated and used to jointly optimize facial expression recognition through a shallow neural network architecture. The joint optimization improves the recognition rates of difficult expressions such as fear and sadness. Furthermore, extensive experiments in both within- and cross-database scenarios have been conducted on relevant benchmark data sets for facial expression recognition: CK+, MMI, JAFFE, and SFEW 2.0. The proposed method (LBF-NN) compares favorably with state-of-the-art algorithms while achieving an order of magnitude improvement in execution time.

Humans can continuously learn new knowledge as their experience grows. In contrast, previous learning in deep neural networks can quickly fade out when they are trained on a new task. In this paper, we hypothesize this problem can be avoided by learning a set of generalized parameters, that are neither specific to old nor new tasks. In this pursuit, we introduce a novel meta-learning approach that seeks to maintain an equilibrium between all the encountered tasks. This is ensured by a new meta-update rule which avoids catastrophic forgetting. In comparison to previous meta-learning techniques, our approach is task-agnostic. When presented with a continuum of data, our model automatically identifies the task and quickly adapts to it with just a single update. We perform extensive experiments on five datasets in a class-incremental setting, leading to significant improvements over the state of the art methods (e.g., a 21.3% boost on CIFAR100 with 10 incremental tasks). Specifically, on large-scale datasets that generally prove difficult cases for incremental learning, our approach delivers absolute gains as high as 19.1% and 7.4% on ImageNet and MS-Celeb datasets, respectively.

In this paper, a novel cognitive architecture for action recognition is developed by applying layers of growing grid neural networks. Using these layers makes the system capable of automatically arranging its representational structure. In addition to the expansion of the neural map during the growth phase, the system is provided with a prior knowledge of the input space, which increases the processing speed of the learning phase. Apart from two layers of growing grid networks the architecture is composed of a preprocessing layer, an ordered vector representation layer and a one-layer supervised neural network. These layers are designed to solve the action recognition problem. The first-layer growing grid receives the input data of human actions and the neural map generates an action pattern vector representing each action sequence by connecting the elicited activation of the trained map. The pattern vectors are then sent to the ordered vector representation layer to build the time-invariant input vectors of key activations for the second-layer growing grid. The second-layer growing grid categorizes the input vectors to the corresponding action clusters/sub-clusters and finally the one-layer supervised neural network labels the shaped clusters with action labels. Three experiments using different datasets of actions show that the system is capable of learning to categorize the actions quickly and efficiently. The performance of the growing grid architecture is compared with the results from a system based on Self-Organizing Maps, showing that the growing grid architecture performs significantly superior on the action recognition tasks.

Adversarial examples can cause catastrophic mistakes in Deep Neural Network (DNNs) based vision systems e.g., for classification, segmentation and object detection. The vulnerability of DNNs against such attacks can prove a major roadblock towards their real-world deployment. Transferability of adversarial examples demand generalizable defenses that can provide cross-task protection. Adversarial training that enhances robustness by modifying target model’s parameters lacks such generalizability. On the other hand, different input processing based defenses fall short in the face of continuously evolving attacks. In this paper, we take the first step to combine the benefits of both approaches and propose a self-supervised adversarial training mechanism in the input space. By design, our defense is a generalizable approach and provides significant robustness against the unseen adversarial attacks (\eg by reducing the success rate of translation-invariant ensemble attack from 82.6% to 31.9% in comparison to previous state-of-the-art). It can be deployed as a plug-and-play solution to protect a variety of vision systems, as we demonstrate for the case of classification, segmentation and detection.

Recent innovations in microelectronic and semiconductor technology enable the creation of smaller and economical hyperspectral cameras. A filter combined camera with advanced scanning module is a game changer that extends the application of miniature hyperspectral imagers to many security domains. This work presents the assessment of the imager L4 from Glana Sensors to detect concealed targets in woodland areas. Several target detection methods were applied to a collection of scenes acquired under various illumination conditions and containing different materials. The potential and limitations of this new imaging device in the context of difficult target detection in forested area are evaluated and discussed.

Generally, convolutional neural networks (CNNs) process data on a regular grid, e.g. data generated by ordinary cameras. Designing CNNs for sparse and irregularly spaced input data is still an open research problem with numerous applications in autonomous driving, robotics, and surveillance. In this paper, we propose an algebraically-constrained normalized convolution layer for CNNs with highly sparse input that has a smaller number of network parameters compared to related work. We propose novel strategies for determining the confidence from the convolution operation and propagating it to consecutive layers. We also propose an objective function that simultaneously minimizes the data error while maximizing the output confidence. To integrate structural information, we also investigate fusion strategies to combine depth and RGB information in our normalized convolution network framework. In addition, we introduce the use of output confidence as an auxiliary information to improve the results. The capabilities of our normalized convolution network framework are demonstrated for the problem of scene depth completion. Comprehensive experiments are performed on the KITTI-Depth and the NYU-Depth-v2 datasets. The results clearly demonstrate that the proposed approach achieves superior performance while requiring only about 1-5% of the number of parameters compared to the state-of-the-art methods.

Probabilistic methods for point set registration have interesting theoretical properties, such as linear complexity in the number of used points, and they easily generalize to joint registration of multiple point sets. In this work, we improve their recognition performance to match state of the art. This is done by incorporating learned features, by adding a von Mises-Fisher feature model in each mixture component, and by using learned attention weights. We learn these jointly using a registration loss learning strategy (RLL) that directly uses the registration error as a loss, by back-propagating through the registration iterations. This is possible as the probabilistic registration is fully differentiable, and the result is a learning framework that is truly end-to-end. We perform extensive experiments on the 3DMatch and Kitti datasets. The experiments demonstrate that our approach benefits significantly from the integration of the learned features and our learning strategy, outperforming the state-of-the-art on Kitti. Code is available at https://github.com/felja633/RLLReg.

Video object segmentation (VOS) is a highly challenging problem since the initial mask, defining the target object, is only given at test-time. The main difficulty is to effectively handle appearance changes and similar background objects, while maintaining accurate segmentation. Most previous approaches fine-tune segmentation networks on the first frame, resulting in impractical frame-rates and risk of overfitting. More recent methods integrate generative target appearance models, but either achieve limited robustness or require large amounts of training data. We propose a novel VOS architecture consisting of two network components. The target appearance model consists of a light-weight module, which is learned during the inference stage using fast optimization techniques to predict a coarse but robust target segmentation. The segmentation model is exclusively trained offline, designed to process the coarse scores into high quality segmentation masks. Our method is fast, easily trainable and remains highly effective in cases of limited training data. We perform extensive experiments on the challenging YouTube-VOS and DAVIS datasets. Our network achieves favorable performance, while operating at higher frame-rates compared to state-of-the-art. Code and trained models are available at https://github.com/andr345/frtm-vos.

The Visual Object Tracking challenge VOT2020 is the eighth annual tracker benchmarking activity organized by the VOT initiative. Results of 58 trackers are presented; many are state-of-the-art trackers published at major computer vision conferences or in journals in the recent years. The VOT2020 challenge was composed of five sub-challenges focusing on different tracking domains: (i) VOT-ST2020 challenge focused on short-term tracking in RGB, (ii) VOT-RT2020 challenge focused on “real-time” short-term tracking in RGB, (iii) VOT-LT2020 focused on long-term tracking namely coping with target disappearance and reappearance, (iv) VOT-RGBT2020 challenge focused on short-term tracking in RGB and thermal imagery and (v) VOT-RGBD2020 challenge focused on long-term tracking in RGB and depth imagery. Only the VOT-ST2020 datasets were refreshed. A significant novelty is introduction of a new VOT short-term tracking evaluation methodology, and introduction of segmentation ground truth in the VOT-ST2020 challenge – bounding boxes will no longer be used in the VOT-ST challenges. A new VOT Python toolkit that implements all these novelites was introduced. Performance of the tested trackers typically by far exceeds standard baselines. The source code for most of the trackers is publicly available from the VOT page. The dataset, the evaluation kit and the results are publicly available at the challenge website (http://votchallenge.net ). 

Deep learning algorithms have improved the speed and quality of segmentation for certain tasks in medical imaging. The aim of this work is to design and evaluate an algorithm capable of segmenting bones in dual-energy CT data sets. A convolutional neural network based on the 3D U-Net architecture was implemented and evaluated using high tube voltage images, mixed images and dual-energy images from 30 patients. The network performed well on all the data sets; the mean Dice coefficient for the test data was larger than 0.963. Of special interest is that it performed better on dual-energy CT volumes compared to mixed images that mimicked images taken at 120 kV. The corresponding increase in the Dice coefficient from 0.965 to 0.966 was small since the enhancements were mainly at the edges of the bones. The method can easily be extended to the segmentation of multi-energy CT data.

Over the last decade, the usage of unmanned systems such as Unmanned Aerial Vehicles (UAVs), Unmanned Surface Vessels (USVs) and Unmanned Ground Vehicles (UGVs) has increased drastically, and there is still a rapid growth. Today, unmanned systems are being deployed in many daily operations, e.g. for deliveries in remote areas, to increase efficiency of agriculture, and for environmental monitoring at sea. For safety reasons, unmanned systems are often the preferred choice for surveillance missions in hazardous environments, e.g. for detection of nuclear radiation, and in disaster areas after earthquakes, hurricanes, or during forest fires. For safe navigation of the unmanned systems during their missions, continuous and accurate global localization and attitude estimation is mandatory.

Over the years, many vision-based methods for position estimation have been developed, primarily for urban areas. In contrast, this thesis is mainly focused on vision-based methods for accurate position and attitude estimates in natural environments, i.e. beyond the urban areas. Vision-based methods possess several characteristics that make them appealing as global position and attitude sensors. First, vision sensors can be realized and tailored for most unmanned vehicle applications. Second, geo-referenced terrain models can be generated worldwide from satellite imagery and can be stored onboard the vehicles. In natural environments, where the availability of geo-referenced images in general is low, registration of image information with terrain models is the natural choice for position and attitude estimation. This is the problem area that I addressed in the contributions of this thesis.

The first contribution is a method for full 6DoF (degrees of freedom) pose estimation from aerial images. A dense local height map is computed using structure from motion. The global pose is inferred from the 3D similarity transform between the local height map and a digital elevation model. Aligning height information is assumed to be more robust to season variations than feature-based matching.

The second contribution is a method for accurate attitude (pitch and roll angle) estimation via horizon detection. It is one of only a few methods that use an omnidirectional (fisheye) camera for horizon detection in aerial images. The method is based on edge detection and a probabilistic Hough voting scheme. The method allows prior knowledge of the attitude angles to be exploited to make the initial attitude estimates more robust. The estimates are then refined through registration with the geometrically expected horizon line from a digital elevation model. To the best of our knowledge, it is the first method where the ray refraction in the atmosphere is taken into account, which enables the highly accurate attitude estimates.

The third contribution is a method for position estimation based on horizon detection in an omnidirectional panoramic image around a surface vessel. Two convolutional neural networks (CNNs) are designed and trained to estimate the camera orientation and to segment the horizon line in the image. The MOSSE correlation filter, normally used in visual object tracking, is adapted to horizon line registration with geometric data from a digital elevation model. Comprehensive field trials conducted in the archipelago demonstrate the GPS-level accuracy of the method, and that the method can be trained on images from one region and then applied to images from a previously unvisited test area.

The CNNs in the third contribution apply the typical scheme of convolutions, activations, and pooling. The fourth contribution focuses on the activations and suggests a new formulation to tune and optimize a piecewise linear activation function during training of CNNs. Improved classification results from experiments when tuning the activation function led to the introduction of a new activation function, the Shifted Exponential Linear Unit (ShELU).

Thermal cameras have historically been of interest mainly for military applications. Increasing image quality and resolution combined with decreasing camera price and size during recent years have, however, opened up new application areas. They are now widely used for civilian applications, e.g., within industry, to search for missing persons, in automotive safety, as well as for medical applications. Thermal cameras are useful as soon as there exists a measurable temperature difference. Compared to cameras operating in the visual spectrum, they are advantageous due to their ability to see in total darkness, robustness to illumination variations, and less intrusion on privacy.

This thesis addresses the problem of automatic image analysis in thermal infrared images with a focus on machine learning methods. The main purpose of this thesis is to study the variations of processing required due to the thermal infrared data modality. In particular, three different problems are addressed: visual object tracking, anomaly detection, and modality transfer. All these are research areas that have been and currently are subject to extensive research. Furthermore, they are all highly relevant for a number of different real-world applications.

The first addressed problem is visual object tracking, a problem for which no prior information other than the initial location of the object is given. The main contribution concerns benchmarking of short-term single-object (STSO) visual object tracking methods in thermal infrared images. The proposed dataset, LTIR (Linköping Thermal Infrared), was integrated in the VOT-TIR2015 challenge, introducing the first ever organized challenge on STSO tracking in thermal infrared video. Another contribution also related to benchmarking is a novel, recursive, method for semi-automatic annotation of multi-modal video sequences. Based on only a few initial annotations, a video object segmentation (VOS) method proposes segmentations for all remaining frames and difficult parts in need for additional manual annotation are automatically detected. The third contribution to the problem of visual object tracking is a template tracking method based on a non-parametric probability density model of the object's thermal radiation using channel representations.

The second addressed problem is anomaly detection, i.e., detection of rare objects or events. The main contribution is a method for truly unsupervised anomaly detection based on Generative Adversarial Networks (GANs). The method employs joint training of the generator and an observation to latent space encoder, enabling stratification of the latent space and, thus, also separation of normal and anomalous samples. The second contribution is the previously unaddressed problem of obstacle detection in front of moving trains using a train-mounted thermal camera. Adaptive correlation filters are updated continuously and missed detections of background are treated as detections of anomalies, or obstacles. The third contribution to the problem of anomaly detection is a method for characterization and classification of automatically detected district heat leakages for the purpose of false alarm reduction.

Finally, the thesis addresses the problem of modality transfer between thermal infrared and visual spectrum images, a previously unaddressed problem. The contribution is a method based on Convolutional Neural Networks (CNNs), enabling perceptually realistic transformations of thermal infrared to visual images. By careful design of the loss function the method becomes robust to image pair misalignments. The method exploits the lower acuity for color differences than for luminance possessed by the human visual system, separating the loss into a luminance and a chrominance part.

Adversarial examples reveal the blind spots of deep neural networks (DNNs) and represent a major concern for security-critical applications. The transferability of adversarial examples makes real-world attacks possible in black-box settings, where the attacker is forbidden to access the internal parameters of the model. The underlying assumption in most adversary generation methods, whether learning an instance-specific or an instance-agnostic perturbation, is the direct or indirect reliance on the original domain-specific data distribution. In this work, for the first time, we demonstrate the existence of domain-invariant adversaries, thereby showing common adversarial space among different datasets and models. To this end, we propose a framework capable of launching highly transferable attacks that crafts adversarial patterns to mislead networks trained on entirely different domains. For instance, an adversarial function learned on Paintings, Cartoons or Medical images can successfully perturb ImageNet samples to fool the classifier, with success rates as high as similar to 99% (l(infinity) &lt;= 10). The core of our proposed adversarial function is a generative network that is trained using a relativistic supervisory signal that enables domain-invariant perturbations. Our approach sets the new state-of-the-art for fooling rates, both under the white-box and black-box scenarios. Furthermore, despite being an instance-agnostic perturbation function, our attack outperforms the conventionally much stronger instance-specific attack methods.

Pedestrian detection relying on deep convolution neural networks has made significant progress. Though promising results have been achieved on standard pedestrians, the performance on heavily occluded pedestrians remains far from satisfactory. The main culprits are intra-class occlusions involving other pedestrians and inter-class occlusions caused by other objects, such as cars and bicycles. These result in a multitude of occlusion patterns. We propose an approach for occluded pedestrian detection with the following contributions. First, we introduce a novel mask-guided attention network that fits naturally into popular pedestrian detection pipelines. Our attention network emphasizes on visible pedestrian regions while suppressing the occluded ones by modulating full body features. Second, we empirically demonstrate that coarse-level segmentation annotations provide reasonable approximation to their dense pixel-wise counterparts. Experiments are performed on CityPersons and Caltech datasets. Our approach sets a new state-of-the-art on both datasets. Our approach obtains an absolute gain of 9.5% in log-average miss rate, compared to the best reported results [31] on the heavily occluded HO pedestrian set of CityPersons test set. Further, on the HO pedestrian set of Caltech dataset, our method achieves an absolute gain of 5.0% in log-average miss rate, compared to the best reported results [13]. Code and models are available at: https://github.com/Leotju/MGAN.

Common object counting in a natural scene is a challenging problem in computer vision with numerous real-world applications. Existing image-level supervised common object counting approaches only predict the global object count and rely on additional instance-level supervision to also determine object locations. We propose an image-level supervised approach that provides both the global object count and the spatial distribution of object instances by constructing an object category density map. Motivated by psychological studies, we further reduce image-level supervision using a limited object count information (up to four). To the best of our knowledge, we are the first to propose image-level supervised density map estimation for common object counting and demonstrate its effectiveness in image-level supervised instance segmentation. Comprehensive experiments are performed on the PASCAL VOC and COCO datasets. Our approach outperforms existing methods, including those using instance-level supervision, on both datasets for common object counting. Moreover, our approach improves state-of-the-art image-level supervised instance segmentation [34] with a relative gain of 17.8% in terms of average best overlap, on the PASCAL VOC 2012 dataset.

This paper presents a method for detecting independently moving objects (IMOs) from a monocular camera mounted on a moving car. We use an existing state of the art monocular sparse visual odometry/SLAM framework, and specifically attack the notorious problem of identifying those IMOs which move parallel to the ego-car motion, that is, in an `epipolar-conformant way. IMO candidate patches are obtained from an existing CNN-based car instance detector. While crossing IMOs can be identified as such by epipolar consistency checks, IMOs that move parallel to the camera motion are much harder to detect as their epipolar conformity allows to misinterpret them as static objects in a wrong distance. We employ a CNN to provide an appearance-based depth estimate, and the ambiguity problem can be solved through depth verification. The obtained motion labels (IMO/static) are then propagated over time using the combination of motion cues and appearance-based information of the IMO candidate patches. We evaluate the performance of our method on the KITTI dataset.

Quantitative dual-energy computed tomography may improve the accuracy of treatment planning in radiation therapy. Of special interest are algorithms that can estimate material composition of the imaged object. One example of such an algorithm is the 2D model-based iterative reconstruction algorithm DIRA. The aim of this work is to extend this algorithm to 3D so that it can be used with cone-beams and helical scanning. In the new algorithm, the parallel FBP method was replaced with the approximate 3D FBP-based PI-method. Its performance was tested using a mathematical phantom consisting of six ellipsoids. The algorithm substantially reduced the beam-hardening artefact and the artefacts caused by approximate reconstruction after six iterations. Compared to Alvarez-Macovskis base material decomposition, DIRA-3D does not require geometrically consistent projections and hence can be used in dual-source CT scanners. Also, it can use several tissue-specific material bases at the same time to represent the imaged object.

Deep learning requires large amounts of annotated data. Manual annotation of objects in video is, regardless of annotation type, a tedious and time-consuming process. In particular, for scarcely used image modalities human annotationis hard to justify. In such cases, semi-automatic annotation provides an acceptable option.

In this work, a recursive, semi-automatic annotation method for video is presented. The proposed method utilizesa state-of-the-art video object segmentation method to propose initial annotations for all frames in a video based on only a few manual object segmentations. In the case of a multi-modal dataset, the multi-modality is exploited to refine the proposed annotations even further. The final tentative annotations are presented to the user for manual correction.

The method is evaluated on a subset of the RGBT-234 visual-thermal dataset reducing the workload for a human annotator with approximately 78% compared to full manual annotation. Utilizing the proposed pipeline, sequences are annotated for the VOT-RGBT 2019 challenge.

Availability of large training datasets was essential for the recent advancement and success of deep learning methods. Due to the difficulties related to biometric data collection, datasets with age and gender annotations are scarce and usually limited in terms of size and sample diversity. Web-scraping approaches for automatic data collection can produce large amounts weakly labeled noisy data. The unsupervised facial biometric data filtering method presented in this paper greatly reduces label noise levels in web-scraped facial biometric data. Experiments on two large state-of-the-art web-scraped facial datasets demonstrate the effectiveness of the proposed method, with respect to training and validation scores, training convergence, and generalization capabilities of trained age and gender estimators.

Generic visual tracking is a challenging computer vision problem, with numerous applications. Most existing approaches rely on appearance information by employing either hand-crafted features or deep RGB features extracted from convolutional neural networks. Despite their success, these approaches struggle in case of ambiguous appearance information, leading to tracking failure. In such cases, we argue that motion cue provides discriminative and complementary information that can improve tracking performance. Contrary to visual tracking, deep motion features have been successfully applied for action recognition and video classification tasks. Typically, the motion features are learned by training a CNN on optical flow images extracted from large amounts of labeled videos. In this paper, we investigate the impact of deep motion features in a tracking-by-detection framework. We also evaluate the fusion of hand-crafted, deep RGB, and deep motion features and show that they contain complementary information. To the best of our knowledge, we are the first to propose fusing appearance information with deep motion features for visual tracking. Comprehensive experiments clearly demonstrate that our fusion approach with deep motion features outperforms standard methods relying on appearance information alone.

Trackers based on discriminative correlation filters (DCF) have recently seen widespread success and in this work we dive into their numerical core. DCF-based trackers interleave learning of the target detector and target state inference based on this detector. Whereas the original formulation includes a closed-form solution for the filter learning, recently introduced improvements to the framework no longer have known closed-form solutions. Instead a large-scale linear least squares problem must be solved each time the detector is updated. We analyze the procedure used to optimize the detector and let the popular scheme introduced with ECO serve as a baseline. The ECO implementation is revisited in detail and several mechanisms are provided with alternatives. With comprehensive experiments we show which configurations are superior in terms of tracking capabilities and optimization performance.

In this work, we address the problem of semi-supervised video object segmentation, where the task is to segment a target object in every image of the video sequence, given a ground truth only in the first frame. To be successful it is crucial to robustly handle unpredictable target appearance changes and distracting objects in the background. In this work we obtain a robust and efficient representation of the target by integrating a fast and light-weight discriminative target model into a deep segmentation network. Trained during inference, the target model learns to discriminate between the local appearances of target and background image regions. Its predictions are enhanced to accurate segmentation masks in a subsequent refinement stage.To further improve the segmentation performance, we add a new module trained to generate global target attention vectors, given the input mask and image feature maps. The attention vectors add semantic information about thetarget from a previous frame to the refinement stage, complementing the predictions provided by the target appearance model. Our method is fast and requires no network fine-tuning. We achieve a combined J and F-score of 70.6 on the DAVIS 2019 test-challenge data

One of the fundamental challenges in video object segmentation is to find an effective representation of the target and background appearance. The best performing approaches resort to extensive fine-tuning of a convolutional neural network for this purpose. Besides being prohibitively expensive, this strategy cannot be truly trained end-to-end since the online fine-tuning procedure is not integrated into the offline training of the network. To address these issues, we propose a network architecture that learns a powerful representation of the target and background appearance in a single forward pass. The introduced appearance module learns a probabilistic generative model of target and background feature distributions. Given a new image, it predicts the posterior class probabilities, providing a highly discriminative cue, which is processed in later network modules. Both the learning and prediction stages of our appearance module are fully differentiable, enabling true end-to-end training of the entire segmentation pipeline. Comprehensive experiments demonstrate the effectiveness of the proposed approach on three video object segmentation benchmarks. We close the gap to approaches based on online fine-tuning on DAVIS17, while operating at 15 FPS on a single GPU. Furthermore, our method outperforms all published approaches on the large-scale YouTube-VOS dataset.

Color description is one of the fundamental problems of image understanding. One of the popular ways to represent colors is by means of color names. Most existing work on color names focuses on only the eleven basic color terms of the English language. This could be limiting the discriminative power of these representations, and representations based on more color names are expected to perform better. However, there exists no clear strategy to choose additional color names. We collect a dataset of 28 additional color names. To ensure that the resulting color representation has high discriminative power we propose a method to order the additional color names according to their complementary nature with the basic color names. This allows us to compute color name representations with high discriminative power of arbitrary length. In the experiments we show that these new color name descriptors outperform the existing color name descriptor on the task of visual tracking, person re-identification and image classification.

Most methods that address computer vision prob-lems require powerful visual features. Many successfulapproaches apply techniques motivated from nonparametricstatistics. The channel representation provides a frameworkfornonparametricdistributionrepresentation.Althoughearlywork has focused on a signal processing view of the rep-resentation, the channel representation can be interpretedin probabilistic terms, e.g., representing the distribution oflocal image orientation. In this paper, a variety of approxi-mative channel-based algorithms for probabilistic problemsare presented: a novel efficient algorithm for density recon-struction, a novel and efficient scheme for nonlinear griddingof densities, and finally a novel method for estimating Copuladensities. The experimental results provide evidence that byrelaxing the requirements for exact solutions, efficient algo-rithms are obtained

We analyze the depth reconstruction precision and sensitivity of two-frame triangulation for the case of general motion, and focus on the case of monocular visual odometry, that is: a single camera looking mostly in the direction of motion. The results confirm intuitive assumptions about the limited triangulation precision close to the focus of expansion.

Selective laser melting (SLM) enables production of highly intricate components. From this point of view, the capabilities of this technology are known to the industry and have been demonstrated in numerous applications. Nonetheless, for serial production purposes the manufacturing industry has so far been reluctant in substituting its conventional methods with SLM. One underlying reason is the lack of simple and reliable process monitoring methods. This study examines the feasibility of using thermography for process monitoring. To this end, an infra-red (IR) camera was mounted off-axis to monitor and record the temperature of every layer. The recorded temperature curves are analysed and interpreted with respect to different stages of the process. Furthermore, the possibility of detecting variations in laser settings by means of thermography is demonstrated. The results show that once thermal patterns are identified, this data can be utilized for in-process and post-process monitoring of SLM production.

In this paper we derive and test a probability-based weighting that can balance residuals of different types in spline fitting. In contrast to previous formulations, the proposed spline error weighting scheme also incorporates a prediction of the approximation error of the spline fit. We demonstrate the effectiveness of the prediction in a synthetic experiment, and apply it to visual-inertial fusion on rolling shutter cameras. This results in a method that can estimate 3D structure with metric scale on generic first-person videos. We also propose a quality measure for spline fitting, that can be used to automatically select the knot spacing. Experiments verify that the obtained trajectory quality corresponds well with the requested quality. Finally, by linearly scaling the weights, we show that the proposed spline error weighting minimizes the estimation errors on real sequences, in terms of scale and end-point errors.

Visual tracking is one of the fundamental problems in computer vision. Its numerous applications include robotics, autonomous driving, augmented reality and 3D reconstruction. In essence, visual tracking can be described as the problem of estimating the trajectory of a target in a sequence of images. The target can be any image region or object of interest. While humans excel at this task, requiring little effort to perform accurate and robust visual tracking, it has proven difficult to automate. It has therefore remained one of the most active research topics in computer vision.

In its most general form, no prior knowledge about the object of interest or environment is given, except for the initial target location. This general form of tracking is known as generic visual tracking. The unconstrained nature of this problem makes it particularly difficult, yet applicable to a wider range of scenarios. As no prior knowledge is given, the tracker must learn an appearance model of the target on-the-fly. Cast as a machine learning problem, it imposes several major challenges which are addressed in this thesis.

The main purpose of this thesis is the study and advancement of the, so called, Discriminative Correlation Filter (DCF) framework, as it has shown to be particularly suitable for the tracking application. By utilizing properties of the Fourier transform, a correlation filter is discriminatively learned by efficiently minimizing a least-squares objective. The resulting filter is then applied to a new image in order to estimate the target location.

This thesis contributes to the advancement of the DCF methodology in several aspects. The main contribution regards the learning of the appearance model: First, the problem of updating the appearance model with new training samples is covered. Efficient update rules and numerical solvers are investigated for this task. Second, the periodic assumption induced by the circular convolution in DCF is countered by proposing a spatial regularization component. Third, an adaptive model of the training set is proposed to alleviate the impact of corrupted or mislabeled training samples. Fourth, a continuous-space formulation of the DCF is introduced, enabling the fusion of multiresolution features and sub-pixel accurate predictions. Finally, the problems of computational complexity and overfitting are addressed by investigating dimensionality reduction techniques.

As a second contribution, different feature representations for tracking are investigated. A particular focus is put on the analysis of color features, which had been largely overlooked in prior tracking research. This thesis also studies the use of deep features in DCF-based tracking. While many vision problems have greatly benefited from the advent of deep learning, it has proven difficult to harvest the power of such representations for tracking. In this thesis it is shown that both shallow and deep layers contribute positively. Furthermore, the problem of fusing their complementary properties is investigated.

The final major contribution of this thesis regards the prediction of the target scale. In many applications, it is essential to track the scale, or size, of the target since it is strongly related to the relative distance. A thorough analysis of how to integrate scale estimation into the DCF framework is performed. A one-dimensional scale filter is proposed, enabling efficient and accurate scale estimation.

Transformation of thermal infrared (TIR) images into visual, i.e. perceptually realistic color (RGB) images, is a challenging problem. TIR cameras have the ability to see in scenarios where vision is severely impaired, for example in total darkness or fog, and they are commonly used, e.g., for surveillance and automotive applications. However, interpretation of TIR images is difficult, especially for untrained operators. Enhancing the TIR image display by transforming it into a plausible, visual, perceptually realistic RGB image presumably facilitates interpretation. Existing grayscale to RGB, so called, colorization methods cannot be applied to TIR images directly since those methods only estimate the chrominance and not the luminance. In the absence of conventional colorization methods, we propose two fully automatic TIR to visual color image transformation methods, a two-step and an integrated approach, based on Convolutional Neural Networks. The methods require neither pre- nor postprocessing, do not require any user input, and are robust to image pair misalignments. We show that the methods do indeed produce perceptually realistic results on publicly available data, which is assessed both qualitatively and quantitatively.

The Exponential Linear Unit (ELU) has been proven to speed up learning and improve the classification performance over activation functions such as ReLU and Leaky ReLU for convolutional neural networks. The reasons behind the improved behavior are that ELU reduces the bias shift, it saturates for large negative inputs and it is continuously differentiable. However, it remains open whether ELU has the optimal shape and we address the quest for a superior activation function.We use a new formulation to tune a piecewise linear activation function during training, to investigate the above question, and learn the shape of the locally optimal activation function. With this tuned activation function, the classification performance is improved and the resulting, learned activation function shows to be ELU-shaped irrespective if it is initialized as a RELU, LReLU or ELU. Interestingly, the learned activation function does not exactly pass through the origin indicating that a shifted ELU-shaped activation function is preferable. This observation leads us to introduce the Shifted Exponential Linear Unit (ShELU) as a new activation function.Experiments on Cifar-100 show that the classification performance is further improved when using the ShELU activation function in comparison with ELU. The improvement is achieved when learning an individual bias shift for each neuron.

We present Lambda Twist; a novel P3P solver which is accurate, fast and robust. Current state-of-the-art P3P solvers find all roots to a quartic and discard geometrically invalid and duplicate solutions in a post-processing step. Instead of solving a quartic, the proposed P3P solver exploits the underlying elliptic equations which can be solved by a fast and numerically accurate diagonalization. This diagonalization requires a single real root of a cubic which is then used to find the, up to four, P3P solutions. Unlike the direct quartic solvers our method never computes geometrically invalid or duplicate solutions.

Extensive evaluation on synthetic data shows that the new solver has better numerical accuracy and is faster compared to the state-of-the-art P3P implementations. Implementation and benchmark are available on github.

This paper investigates the problem of position estimation of unmanned surface vessels (USVs) operating in coastal areas or in the archipelago. We propose a position estimation method where the horizon line is extracted in a 360 degree panoramic image around the USV. We design a CNN architecture to determine an approximate horizon line in the image and implicitly determine the camera orientation (the pitch and roll angles). The panoramic image is warped to compensate for the camera orientation and to generate an image from an approximately level camera. A second CNN architecture is designed to extract the pixelwise horizon line in the warped image. The extracted horizon line is correlated with digital elevation model (DEM) data in the Fourier domain using a MOSSE correlation filter. Finally, we determine the location of the maximum correlation score over the search area to estimate the position of the USV. Comprehensive experiments are performed in a field trial in the archipelago. Our approach provides promising results by achieving position estimates with GPS-level accuracy.

In the field of generic object tracking numerous attempts have been made to exploit deep features. Despite all expectations, deep trackers are yet to reach an outstanding level of performance compared to methods solely based on handcrafted features. In this paper, we investigate this key issue and propose an approach to unlock the true potential of deep features for tracking. We systematically study the characteristics of both deep and shallow features, and their relation to tracking accuracy and robustness. We identify the limited data and low spatial resolution as the main challenges, and propose strategies to counter these issues when integrating deep features for tracking. Furthermore, we propose a novel adaptive fusion approach that leverages the complementary properties of deep and shallow features to improve both robustness and accuracy. Extensive experiments are performed on four challenging datasets. On VOT2017, our approach significantly outperforms the top performing tracker from the challenge with a relative gain of >17% in EAO.

Under the title "Probabilistic and Biologically Inspired Feature Representations," this text collects a substantial amount of work on the topic of channel representations. Channel representations are a biologically motivated, wavelet-like approach to visual feature descriptors: they are local and compact, they form a computational framework, and the represented information can be reconstructed. The first property is shared with many histogram- and signature-based descriptors, the latter property with the related concept of population codes. In their unique combination of properties, channel representations become a visual Swiss army knife—they can be used for image enhancement, visual object tracking, as 2D and 3D descriptors, and for pose estimation. In the chapters of this text, the framework of channel representations will be introduced and its attributes will be elaborated, as well as further insight into its probabilistic modeling and algorithmic implementation will be given. Channel representations are a useful toolbox to represent visual information for machine learning, as they establish a generic way to compute popular descriptors such as HOG, SIFT, and SHOT. Even in an age of deep learning, they provide a good compromise between hand-designed descriptors and a-priori structureless feature spaces as seen in the layers of deep networks.

Privacy protection may be defined as replacing the original content in an image region with a new (less intrusive) content having modified target appearance information to make it less recognizable by applying a privacy protection technique. Indeed the development of privacy protection techniques needs also to be complemented with an established objective evaluation method to facilitate their assessment and comparison. Generally, existing evaluation methods rely on the use of subjective judgements or assume a specific target type in image data and use target detection and recognition accuracies to assess privacy protection. This work proposes a new annotation-free evaluation method that is neither subjective nor assumes a specific target type. It assesses two key aspects of privacy protection: protection and utility. Protection is quantified as an appearance similarity and utility is measured as a structural similarity between original and privacy-protected image regions. We performed an extensive experimentation using six challenging datasets (having 12 video sequences) including a new dataset (having six sequences) that contains visible and thermal imagery. The new dataset, called TST-Priv, is made available online below for community. We demonstrate effectiveness of the proposed method by evaluating six image-based privacy protection techniques, and also show comparisons of the proposed method over existing methods.

Hyperspectral remote sensing based on unmanned airborne vehicles is a field increasing in importance. The combined functionality of simultaneous hyperspectral and geometric modeling is less developed. A configuration has been developed that enables the reconstruction of the hyperspectral three-dimensional (3D) environment. The hyperspectral camera is based on a linear variable filter and a high frame rate, high resolution camera enabling point-to-point matching and 3D reconstruction. This allows the information to be combined into a single and complete 3D hyperspectral model. In this paper, we describe the camera and illustrate capabilities and difficulties through real-world experiments.

Visual odometry using only a monocular camera faces more algorithmic challenges than stereo odometry. We present a robust monocular visual odometry framework for automotive applications. An extended propagation-based tracking framework is proposed which yields highly accurate (unscaled) pose estimates. Scale is supplied by ground plane pose estimation employing street pixel labeling using a convolutional neural network (CNN). The proposed framework has been extensively tested on the KITTI dataset and achieves a higher rank than current published state-of-the-art monocular methods in the KITTI odometry benchmark. Unlike other VO/SLAM methods, this result is achieved without loop closing mechanism, without RANSAC and also without multiframe bundle adjustment. Thus, we challenge the common belief that robust systems can only be built using iterative robustification tools like RANSAC. (C) 2017 Published by Elsevier B.V.

Most iterative optimization algorithms for motion, depth estimation or scene reconstruction, both sparse and dense, rely on a coarse and reliable dense initialization to bootstrap their optimization procedure. This makes techniques important that allow to obtain a dense but still approximative representation of a desired 2D structure (e.g., depth maps, optical flow, disparity maps) from a very sparse measurement of this structure. The method presented here exploits the complete information given by the principal component analysis (PCA), the principal basis and its prior distribution. The method is able to determine a dense reconstruction even if only a very sparse measurement is available. When facing such situations, typically the number of principal components is further reduced which results in a loss of expressiveness of the basis. We overcome this problem and inject prior knowledge in a maximum a posteriori (MAP) approach. We test our approach on the KITTI and the Virtual KITTI dataset and focus on the interpolation of depth maps for driving scenes. The evaluation of the results shows good agreement to the ground truth and is clearly superior to the results of an interpolation by the nearest neighbor method which disregards statistical information.

The recognition of individual object instances in single monocular images is still an incompletely solved task. In this work, we propose a new approach for detecting and separating vehicles in the context of autonomous driving. Our method uses the fully convolutional network (FCN) for semantic labeling and for estimating the boundary of each vehicle. Even though a contour is in general a one pixel wide structure which cannot be directly learned by a CNN, our network addresses this by providing areas around the contours. Based on these areas, we separate the individual vehicle instances. In our experiments, we show on two challenging datasets (Cityscapes and KITTI) that we achieve state-of-the-art performance, despite the usage of a subsampling rate of two. Our approach even outperforms all recent works w.r.t. several rating scores.

Object recognition is a skill we as humans often take for granted. Due to our formidable object learning, recognition and generalisation skills, it is sometimes hard to see the multitude of obstacles that need to be overcome in order to replicate this skill in an artificial system. Object recognition is also one of the classical areas of computer vision, and many ways of approaching the problem have been proposed. Recently, visually capable robots and autonomous vehicles have increased the focus on embodied recognition systems and active visual search. These applications demand that systems can learn and adapt to their surroundings, and arrive at decisions in a reasonable amount of time, while maintaining high object recognition performance. This is especially challenging due to the high dimensionality of image data. In cases where end-to-end learning from pixels to output is needed, mechanisms designed to make inputs tractable are often necessary for less computationally capable embodied systems.Active visual search also means that mechanisms for attention and gaze control are integral to the object recognition procedure. Therefore, the way in which attention mechanisms should be introduced into feature extraction and estimation algorithms must be carefully considered when constructing a recognition system.This thesis describes work done on the components necessary for creating an embodied recognition system, specifically in the areas of decision uncertainty estimation, object segmentation from multiple cues, adaptation of stereo vision to a specific platform and setting, problem-specific feature selection, efficient estimator training and attentional modulation in convolutional neural networks. Contributions include the evaluation of methods and measures for predicting the potential uncertainty reduction that can be obtained from additional views of an object, allowing for adaptive target observations. Also, in order to separate a specific object from other parts of a scene, it is often necessary to combine multiple cues such as colour and depth in order to obtain satisfactory results. Therefore, a method for combining these using channel coding has been evaluated. In order to make use of three-dimensional spatial structure in recognition, a novel stereo vision algorithm extension along with a framework for automatic stereo tuning have also been investigated. Feature selection and efficient discriminant sampling for decision tree-based estimators have also been implemented. Finally, attentional multi-layer modulation of convolutional neural networks for recognition in cluttered scenes has been evaluated. Several of these components have been tested and evaluated on a purpose-built embodied recognition platform known as Eddie the Embodied.

More and more devices have depth sensors, making RGB+D video (colour+depth video) increasingly common. RGB+D video allows the use of depth image based rendering (DIBR) to render a given scene from different viewpoints, thus making it a useful asset in view prediction for 3D and free-viewpoint video coding. In this paper we evaluate a multitude of algorithms for scattered data interpolation, in order to optimize the performance of DIBR for video coding. This also includes novel contributions like a Kriging refinement step, an edge suppression step to suppress artifacts, and a scale-adaptive kernel. Our evaluation uses the depth extension of the Sintel datasets. Using ground-truth sequences is crucial for such an optimization, as it ensures that all errors and artifacts are caused by the prediction itself rather than noisy or erroneous data. We also present a comparison with the commonly used mesh-based projection.

The ability to direct visual attention is a fundamental skill for seeing robots. Attention comes in two flavours: the gaze direction (overt attention) and attention to a specific part of the current field of view (covert attention), of which the latter is the focus of the present study. Specifically, we study the effects of attentional masking within pre-trained deep neural networks for the purpose of handling ambiguous scenes containing multiple objects. We investigate several variants of attentional masking on partially pre-trained deep neural networks and evaluate the effects on classification performance and sensitivity to attention mask errors in multi-object scenes. We find that a combined scheme consisting of multi-level masking and blending provides the best trade-off between classification accuracy and insensitivity to masking errors. This proposed approach is denoted multilayer continuous-valued convolutional feature masking (MC-CFM). For reasonably accurate masks it can suppress the influence of distracting objects and reach comparable classification performance to unmasked recognition in cases without distractors.

An increasing number of robots and autonomous vehicles are equipped with multiple cameras to achieve surround-view sensing. The estimation of their relative poses, also known as extrinsic parameter calibration, is a challenging problem, particularly in the non-overlapping case. We present a simple and novel extrinsic calibration method based on standard components that performs favorably to existing approaches. We further propose a framework for predicting the performance of different calibration configurations and intuitive error metrics. This makes selecting a good camera configuration straightforward. We evaluate on rendered synthetic images and show good results as measured by angular and absolute pose differences, as well as the reprojection error distributions.

Accurate scale estimation of a target is a challenging research problem in visual object tracking. Most state-of-the-art methods employ an exhaustive scale search to estimate the target size. The exhaustive search strategy is computationally expensive and struggles when encountered with large scale variations. This paper investigates the problem of accurate and robust scale estimation in a tracking-by-detection framework. We propose a novel scale adaptive tracking approach by learning separate discriminative correlation filters for translation and scale estimation. The explicit scale filter is learned online using the target appearance sampled at a set of different scales. Contrary to standard approaches, our method directly learns the appearance change induced by variations in the target scale. Additionally, we investigate strategies to reduce the computational cost of our approach. Extensive experiments are performed on the OTB and the VOT2014 datasets. Compared to the standard exhaustive scale search, our approach achieves a gain of 2.5 percent in average overlap precision on the OTB dataset. Additionally, our method is computationally efficient, operating at a 50 percent higher frame rate compared to the exhaustive scale search. Our method obtains the top rank in performance by outperforming 19 state-of-the-art trackers on OTB and 37 state-of-the-art trackers on VOT2014.

Semantic segmentation of 3D point clouds is a challenging problem with numerous real-world applications. While deep learning has revolutionized the field of image semantic segmentation, its impact on point cloud data has been limited so far. Recent attempts, based on 3D deep learning approaches (3D-CNNs), have achieved below-expected results. Such methods require voxelizations of the underlying point cloud data, leading to decreased spatial resolution and increased memory consumption. Additionally, 3D-CNNs greatly suffer from the limited availability of annotated datasets.

Research in optical flow estimation has to a large extent focused on achieving the best possible quality with no regards to running time. Nevertheless, in a number of important applications the speed is crucial. To address this problem we present BriefMatch, a real-time optical flow method that is suitable for live applications. The method combines binary features with the search strategy from PatchMatch in order to efficiently find a dense correspondence field between images. We show that the BRIEF descriptor provides better candidates (less outlier-prone) in shorter time, when compared to direct pixel comparisons and the Census transform. This allows us to achieve high quality results from a simple filtering of the initially matched candidates. Currently, BriefMatch has the fastest running time on the Middlebury benchmark, while placing highest of all the methods that run in shorter than 0.5 seconds.

We, the Guest Editors of this special issue of Pattern Recognition Letters are pleased to share these contributions with you. The papers included here are based on work from the 22nd International Conference on Pattern Recognition (IAPR) in Stockholm, Sweden, held August 24–28, 2014. The papers selected for this special issue were those winning one of the IAPR awards, as well as one paper by a former student of the winner of the KS Fu Prize, Prof. Jitendra Malik. Taken together, this body of work represents some of the finest research being conducted by the IAPR community worldwide, it builds on a rich legacy of accomplishment by the entire community, and it offers a view to the future, to where we are going as a scientific community.

For each of the award-winning papers, the authors were asked to revise and extend their contributions to full journal length and to provide true added value vis-à-vis the original conference submission. In some cases, the authors elected to modify the titles slightly, and in some cases the list of authors has also been modified. The resulting manuscripts were sent out for full review by a different set of referees than those who reviewed the conference versions. The process, including required revisions, was in accordance with the standing editorial policy of Pattern Recognition Letters, resulting in the final versions accepted and appearing here. These are thoroughly vetted, high-caliber scientific contributions.

It has been our honor to serve as Guest Editors for this special issue. We would like to thank the Editors of Pattern Recognition Letters for allowing us this opportunity. We are especially grateful to Dr. Gabriella Sanniti di Baja for her enthusiasm, support, and her willingness to keep prodding us along to bring the special issue through to completion. We would also like to thank all of those who reviewed the papers, both originally for the conference and subsequently for the journal, and those who served on the ICPR awards and KS Fu Prize committees.

Finally, we express our heartfelt gratitude to all of the authors for taking the time to prepare these versions for our collective enlightenment, sharing their knowledge, innovation, and discoveries with the rest of us.

Phase correlation is one of the classic methods for sparse motion or displacement estimation. It is renowned in the literature for high precision and insensitivity against illumination variations. We propose several important enhancements to the phase correlation (PhC) method which render it more robust against those situations where a motion measurement is not possible (low structure, too much noise, too different image content in the corresponding measurement windows). This allows the method to perform self-diagnosis in adverse situations. Furthermore, we extend the PhC method by a robust scheme for detecting and classifying the presence of multiple motions and estimating their uncertainties. Experimental results on the Middlebury Stereo Dataset and on the KITTI Optical Flow Dataset show the potential offered by the enhanced method in contrast to the PhC implementation of OpenCV.

Current best local descriptors are learned on a large dataset of matching and non-matching keypoint pairs. However, data of this kind is not always available since detailed keypoint correspondences can be hard to establish. On the other hand, we can often obtain labels for pairs of keypoint bags. For example, keypoint bags extracted from two images of the same object under different views form a matching pair, and keypoint bags extracted from images of different objects form a non-matching pair. On average, matching pairs should contain more corresponding keypoints than non-matching pairs. We describe an end-to-end differentiable architecture that enables the learning of local keypoint descriptors from such weakly-labeled data.

One of the major steps in visual environment perception for automotive applications is to track keypoints and to subsequently estimate egomotion and environment structure from the trajectories of these keypoints. This paper presents a propagation based tracking method to obtain the 2D trajectories of keypoints from a sequence of images in a monocular camera setup. Instead of relying on the classical RANSAC to obtain accurate keypoint correspondences, we steer the search for keypoint matches by means of propagating the estimated 3D position of the keypoint into the next frame and verifying the photometric consistency. In this process, we continuously predict, estimate and refine the frame-to-frame relative pose which induces the epipolar relation. Experiments on the KITTI dataset as well as on the synthetic COnGRATS dataset show promising results on the estimated courses and accurate keypoint trajectories.

Correspondence relations between different views of the same scene can be learnt in an unsupervised manner. We address autonomous learning of arbitrary fixed spatial (point-to-point) mappings. Since any such transformation can be represented by a permutation matrix, the signal model is a linear one, whereas the proposed analysis method, mainly based on Canonical Correlation Analysis (CCA) is based on a generalized eigensystem problem, i.e., a nonlinear operation. The learnt transformation is represented implicitly in terms of pairs of learned basis vectors and does neither use nor require an analytic/parametric expression for the latent mapping. We show how the rank of the signal that is shared among views may be determined from canonical correlations and how the overlapping (=shared) dimensions among the views may be inferred.

Attitude (pitch and roll angle) estimation from visual information is necessary for GPS-free navigation of airborne vehicles. We propose a highly accurate method to estimate the attitude by horizon detection in fisheye images. A Canny edge detector and a probabilistic Hough voting scheme are used to compute an approximate attitude and the corresponding horizon line in the image. Horizon edge pixels are extracted in a band close to the approximate horizon line. The attitude estimates are refined through registration of the extracted edge pixels with the geometrical horizon from a digital elevation map (DEM), in our case the SRTM3 database, extracted at a given approximate position. The proposed method has been evaluated using 1629 images from a flight trial with flight altitudes up to 600 m in an area with ground elevations ranging from sea level up to 500 m. Compared with the ground truth from a filtered inertial measurement unit (IMU)/GPS solution, the standard deviation for the pitch and roll angle errors obtained with 30 Mpixel images are 0.04° and 0.05°, respectively, with mean errors smaller than 0.02°. To achieve the high-accuracy attitude estimates, the ray refraction in the earth's atmosphere has been taken into account. The attitude errors obtained on real images are less or equal to those achieved on synthetic images for previous methods with DEM refinement, and the errors are about one order of magnitude smaller than for any previous vision-based method without DEM refinement.

Driver assistance systems in modern cars now show clear steps towards autonomous driving and improvements are presented in a steady pace. The total number of sensors has also decreased from the vehicles of the initial DARPA challenge, more resembling a pile of sensors with a car underneath. Still, anyone driving a tele-operated toy using a video link is a demonstration that a single camera provides enough information about the surronding world.  

Most lane assist systems are developed for highway use and depend on visible lane markers. However, lane markers may not be visible due to snow or wear, and there are roads without lane markers. With a slightly different approach, autonomous road following can be obtained on almost any kind of road. Using realtime online machine learning, a human driver can demonstrate driving on a road type unknown to the system and after some training, the system can seamlessly take over. The demonstrator system presented in this work has shown capability of learning to follow different types of roads as well as learning to follow a person. The system is based solely on vision, mapping camera images directly to control signals.  

Such systems need the ability to handle multiple-hypothesis outputs as there may be several plausible options in similar situations. If there is an obstacle in the middle of the road, the obstacle can be avoided by going on either side. However the average action, going straight ahead, is not a viable option. Similarly, at an intersection, the system should follow one road, not the average of all roads.  

To this end, an online machine learning framework is presented where inputs and outputs are represented using the channel representation. The learning system is structurally simple and computationally light, based on neuropsychological ideas presented by Donald Hebb over 60 years ago. Nonetheless the system has shown a cabability to learn advanced tasks. Furthermore, the structure of the system permits a statistical interpretation where a non-parametric representation of the joint distribution of input and output is generated. Prediction generates the conditional distribution of the output, given the input.  

The statistical interpretation motivates the introduction of priors. In cases with multiple options, such as at intersections, a prior can select one mode in the multimodal distribution of possible actions. In addition to the ability to learn from demonstration, a possibility for immediate reinforcement feedback is presented. This allows for a system where the teacher can choose the most appropriate way of training the system, at any time and at her own discretion.  

The theoretical contributions include a deeper analysis of the channel representation. A geometrical analysis illustrates the cause of decoding bias commonly present in neurologically inspired representations, and measures to counteract it. Confidence values are analyzed and interpreted as evidence and coherence. Further, the use of the truncated cosine basis function is motivated.  

Finally, a selection of applications is presented, such as autonomous road following by online learning and head pose estimation. A method founded on the same basic principles is used for visual tracking, where the probabilistic representation of target pixel values allows for changes in target appearance.

Recent years have shown great progress in driving assistance systems, approaching autonomous driving step by step. Many approaches rely on lane markers however, which limits the system to larger paved roads and poses problems during winter. In this work we explore an alternative approach to visual road following based on online learning. The system learns the current visual appearance of the road while the vehicle is operated by a human. When driving onto a new type of road, the human driver will drive for a minute while the system learns. After training, the human driver can let go of the controls. The present work proposes a novel approach to online perception-action learning for the specific problem of road following, which makes interchangeably use of supervised learning (by demonstration), instantaneous reinforcement learning, and unsupervised learning (self-reinforcement learning). The proposed method, symbiotic online learning of associations and regression (SOLAR), extends previous work on qHebb-learning in three ways: priors are introduced to enforce mode selection and to drive learning towards particular goals, the qHebb-learning methods is complemented with a reinforcement variant, and a self-assessment method based on predictive coding is proposed. The SOLAR algorithm is compared to qHebb-learning and deep learning for the task of road following, implemented on a model RC-car. The system demonstrates an ability to learn to follow paved and gravel roads outdoors. Further, the system is evaluated in a controlled indoor environment which provides quantifiable results. The experiments show that the SOLAR algorithm results in autonomous capabilities that go beyond those of existing methods with respect to speed, accuracy, and functionality. 

Advanced model-based iterative reconstruction algorithms in quantitative computed tomography (CT) perform automatic segmentation of tissues to estimate material properties of the imaged object. Compared with conventional methods, these algorithms may improve quality of reconstructed images and accuracy of radiation treatment planning. Automatic segmentation of tissues is, however, a difficult task. The aim of this work was to develop and evaluate an algorithm that automatically segments tissues in CT images of the male pelvis. The newly developed algorithm (MK2014) combines histogram matching, thresholding, region growing, deformable model and atlas-based registration techniques for the segmentation of bones, adipose tissue, prostate and muscles in CT images. Visual inspection of segmented images showed that the algorithm performed well for the five analysed images. The tissues were identified and outlined with accuracy sufficient for the dual-energy iterative reconstruction algorithm whose aim is to improve the accuracy of radiation treatment planning in brachytherapy of the prostate.

The Thermal Infrared Visual Object Tracking (VOT-TIR2015) Challenge was organized in conjunction with ICCV2015. It was the first benchmark on short-term,single-target tracking in thermal infrared (TIR) sequences. The challenge aimed at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. It was based on the VOT2013 Challenge, but introduced the following novelties: (i) the utilization of the LTIR (Linköping TIR) dataset, (ii) adaption of the VOT2013 attributes to thermal data, (iii) a similar evaluation to that of VOT2015. This paper provides an overview of the VOT-TIR2015 Challenge as well as the results of the 24 participating trackers.

The Visual Object Tracking challenge VOT2016 aims at comparing short-term single-object visual trackers that do not apply pre-learned models of object appearance. Results of 70 trackers are presented, with a large number of trackers being published at major computer vision conferences and journals in the recent years. The number of tested state-of-the-art trackers makes the VOT 2016 the largest and most challenging benchmark on short-term tracking to date. For each participating tracker, a short description is provided in the Appendix. The VOT2016 goes beyond its predecessors by (i) introducing a new semi-automatic ground truth bounding box annotation methodology and (ii) extending the evaluation system with the no-reset experiment.

Tracking-by-detection methods have demonstrated competitive performance in recent years. In these approaches, the tracking model heavily relies on the quality of the training set. Due to the limited amount of labeled training data, additional samples need to be extracted and labeled by the tracker itself. This often leads to the inclusion of corrupted training samples, due to occlusions, misalignments and other perturbations. Existing tracking-by-detection methods either ignore this problem, or employ a separate component for managing the training set. We propose a novel generic approach for alleviating the problem of corrupted training samples in tracking-by-detection frameworks. Our approach dynamically manages the training set by estimating the quality of the samples. Contrary to existing approaches, we propose a unified formulation by minimizing a single loss over both the target appearance model and the sample quality weights. The joint formulation enables corrupted samples to be down-weighted while increasing the impact of correct ones. Experiments are performed on three benchmarks: OTB-2015 with 100 videos, VOT-2015 with 60 videos, and Temple-Color with 128 videos. On the OTB-2015, our unified formulation significantly improves the baseline, with a gain of 3.8% in mean overlap precision. Finally, our method achieves state-of-the-art results on all three datasets.

3D-point set registration is an active area of research in computer vision. In recent years, probabilistic registration approaches have demonstrated superior performance for many challenging applications. Generally, these probabilistic approaches rely on the spatial distribution of the 3D-points, and only recently color information has been integrated into such a framework, significantly improving registration accuracy. Other than local color information, high-dimensional 3D shape features have been successfully employed in many applications such as action recognition and 3D object recognition. In this paper, we propose a probabilistic framework to integrate high-dimensional 3D shape features with color information for point set registration. The 3D shape features are distinctive and provide complementary information beneficial for robust registration. We validate our proposed framework by performing comprehensive experiments on the challenging Stanford Lounge dataset, acquired by a RGB-D sensor, and an outdoor dataset captured by a Lidar sensor. The results clearly demonstrate that our approach provides superior results both in terms of robustness and accuracy compared to state-of-the-art probabilistic methods.

The Thermal Infrared Visual Object Tracking challenge 2016, VOT-TIR2016, aims at comparing short-term single-object visual trackers that work on thermal infrared (TIR) sequences and do not apply pre-learned models of object appearance. VOT-TIR2016 is the second benchmark on short-term tracking in TIR sequences. Results of 24 trackers are presented. For each participating tracker, a short description is provided in the appendix. The VOT-TIR2016 challenge is similar to the 2015 challenge, the main difference is the introduction of new, more difficult sequences into the dataset. Furthermore, VOT-TIR2016 evaluation adopted the improvements regarding overlap calculation in VOT2016. Compared to VOT-TIR2015, a significant general improvement of results has been observed, which partly compensate for the more difficult sequences. The dataset, the evaluation kit, as well as the results are publicly available at the challenge website.

1 Migratory songbirds carry an inherited capacity to migrate several thousand kilometers each year crossing continental landmasses and barriers between distant breeding sites and wintering areas. How individual songbirds manage with extreme precision to find their way is still largely unknown. The functional characteristics of biological compasses used by songbird migrants has mainly been investigated by recording the birds directed migratory activity in circular cages, so-called Emlen funnels. This method is 50 years old and has not received major updates over the past decades. The aim of this work was to compare the results from newly developed digital methods with the established manual methods to evaluate songbird migratory activity and orientation in circular cages. 2 We performed orientation experiments using the European robin (Erithacus rubecula) using modified Emlen funnels equipped with thermal paper and simultaneously recorded the songbird movements from above. We evaluated and compared the results obtained with five different methods. Two methods have been commonly used in songbirds orientation experiments; the other three methods were developed for this study and were based either on evaluation of the thermal paper using automated image analysis, or on the analysis of videos recorded during the experiment. 3 The methods used to evaluate scratches produced by the claws of birds on the thermal papers presented some differences compared with the video analyses. These differences were caused mainly by differences in scatter, as any movement of the bird along the sloping walls of the funnel was recorded on the thermal paper, whereas video evaluations allowed us to detect single takeoff attempts by the birds and to consider only this behavior in the orientation analyses. Using computer vision, we were also able to identify and separately evaluate different behaviors that were impossible to record by the thermal paper. 4 The traditional Emlen funnel is still the most used method to investigate compass orientation in songbirds under controlled conditions. However, new numerical image analysis techniques provide a much higher level of detail of songbirds migratory behavior and will provide an increasing number of possibilities to evaluate and quantify specific behaviors as new algorithms will be developed.

Many image processing methods such as corner detection,optical flow and iterative enhancement make use of image tensors. Generally, these tensors are estimated using the structure tensor. In this work we show that the gradient energy tensor can be used as an alternativeto the structure tensor in several cases. We apply the gradient energy tensor to common image problem applications such as corner detection, optical flow and image enhancement. Our experimental results suggest that the gradient energy tensor enables real-time tensor-based image enhancement using the graphical processing unit (GPU) and we obtain 40% increase of frame rate without loss of image quality.

Action recognition in still images is a challenging problem in computer vision. To facilitate comparative evaluation independently of person detection, the standard evaluation protocol for action recognition uses an oracle person detector to obtain perfect bounding box information at both training and test time. The assumption is that, in practice, a general person detector will provide candidate bounding boxes for action recognition. In this paper, we argue that this paradigm is suboptimal and that action class labels should already be considered during the detection stage. Motivated by the observation that body pose is strongly conditioned on action class, we show that: 1) the existing state-of-the-art generic person detectors are not adequate for proposing candidate bounding boxes for action classification; 2) due to limited training examples, the direct training of action-specific person detectors is also inadequate; and 3) using only a small number of labeled action examples, the transfer learning is able to adapt an existing detector to propose higher quality bounding boxes for subsequent action classification. To the best of our knowledge, we are the first to investigate transfer learning for the task of action-specific person detection in still images. We perform extensive experiments on two benchmark data sets: 1) Stanford-40 and 2) PASCAL VOC 2012. For the action detection task (i.e., both person localization and classification of the action performed), our approach outperforms methods based on general person detection by 5.7% mean average precision (MAP) on Stanford-40 and 2.1% MAP on PASCAL VOC 2012. Our approach also significantly outperforms the state of the art with a MAP of 45.4% on Stanford-40 and 31.4% on PASCAL VOC 2012. We also evaluate our action detection approach for the task of action classification (i.e., recognizing actions without localizing them). For this task, our approach, without using any ground-truth person localization at test time, outperforms on both data sets state-of-the-art methods, which do use person locations.

In recent years, short-term single-object tracking has emerged has a popular research topic, as it constitutes the core of more general tracking systems. Many such tracking methods are based on matching a part of the image with a template that is learnt online and represented by, for example, a correlation filter or a distribution field. In order for such a tracker to be able to not only find the position, but also the scale, of the tracked object in the next frame, some kind of scale estimation step is needed. This step is sometimes separate from the position estimation step, but is nevertheless jointly evaluated in de facto benchmarks. However, for practical as well as scientific reasons, the scale estimation step should be evaluated separately – for example,theremightincertainsituationsbeothermethodsmore suitable for the task. In this paper, we describe an evaluation method for scale estimation in template-based short-term single-object tracking, and evaluate two state-of-the-art tracking methods where estimation of scale and position are separable.

The motion of a driving car is highly constrained and we claim that powerful predictors can be built that learn the typical egomotion statistics, and support the typical tasks of feature matching, tracking, and egomotion estimation. We analyze the statistics of the ground truth data given in the KITTI odometry benchmark sequences and confirm that a coordinated turn motion model, overlaid by moderate vibrations, is a very realistic model. We develop a predictor that is able to significantly reduce the uncertainty about the relative motion when a new image frame comes in. Such predictors can be used to steer the matching process from frame n to frame n + 1. We show that they can also be employed to detect outliers in the temporal sequence of egomotion parameters.

During recent years, thermal cameras have decreased in both size and cost while improving image quality. The area of use for such cameras has expanded with many exciting applications, many of which require tracking of objects. While being subject to extensive research in the visual domain, tracking in thermal imagery has historically been of interest mainly for military purposes. The available thermal infrared datasets for evaluating methods addressing these problems are few and the ones that do are not challenging enough for today’s tracking algorithms. Therefore, we hereby propose a thermal infrared dataset for evaluation of short-term tracking methods. The dataset consists of 20 sequences which have been collected from multiple sources and the data format used is in accordance with the Visual Object Tracking (VOT) Challenge.

Describing textures is a challenging problem in computer vision and pattern recognition. The classification problem involves assigning a category label to the texture class it belongs to. Several factors such as variations in scale, illumination and viewpoint make the problem of texture description extremely challenging. A variety of histogram based texture representations exists in literature. However, combining multiple texture descriptors and assessing their complementarity is still an open research problem. In this paper, we first show that combining multiple local texture descriptors significantly improves the recognition performance compared to using a single best method alone. This gain in performance is achieved at the cost of high-dimensional final image representation. To counter this problem, we propose to use an information-theoretic compression technique to obtain a compact texture description without any significant loss in accuracy. In addition, we perform a comprehensive evaluation of pure color descriptors, popular in object recognition, for the problem of texture classification. Experiments are performed on four challenging texture datasets namely, KTH-TIPS-2a, KTH-TIPS-2b, FMD and Texture-10. The experiments clearly demonstrate that our proposed compact multi-texture approach outperforms the single best texture method alone. In all cases, discriminative color names outperforms other color features for texture classification. Finally, we show that combining discriminative color names with compact texture representation outperforms state-of-the-art methods by 7.8%,4.3%7.8%,4.3% and 5.0%5.0% on KTH-TIPS-2a, KTH-TIPS-2b and Texture-10 datasets respectively.