Search

As natural images usually contain multiple objects, multi-label image classification is more applicable "in the wild" than single-label classification. However, exhaustively annotating images with every object of interest is costly and time-consuming. We aim to train multi-label classifiers from single-label annotations only. We show that adding a consistency loss, ensuring that the predictions of the network are consistent over consecutive training epochs, is a simple yet effective method to train multi-label classifiers in a weakly supervised setting. We further extend this approach spatially, by ensuring consistency of the spatial feature maps produced over consecutive training epochs, maintaining per-class running-average heatmaps for each training image. We show that this spatial consistency loss further improves the multi-label mAP of the classifiers. In addition, we show that this method overcomes shortcomings of the "crop" data-augmentation by recovering correct supervision signal even when most of the single ground truth object is cropped out of the input image by the data augmentation. We demonstrate gains of the consistency and spatial consistency losses over the binary cross-entropy baseline, and over competing methods, on MS-COCO and Pascal VOC. We also demonstrate improved multi-label classification mAP on ImageNet-1K using the ReaL multi-label validation set., Comment: 24 pages, 9 figures

ispartof: pages:269-278 ispartof: 2023 IEEE/CVF Winter Conference on Applications of Computer Vision (WACV) pages:269-278 ispartof: IEEE/CVF winter conference on applications of computer vision - WACV 2023 location:Waikoloa, Hawaii, USA date:3 Jan - 7 Jan 2023 status: published

Funder: International Brain Research Organization (IBRO); doi: https://doi.org/10.13039/501100001675, Incrementally learning new information from a non-stationary stream of data, referred to as 'continual learning', is a key feature of natural intelligence, but a challenging problem for deep neural networks. In recent years, numerous deep learning methods for continual learning have been proposed, but comparing their performances is difficult due to the lack of a common framework. To help address this, we describe three fundamental types, or 'scenarios', of continual learning: task-incremental, domain-incremental and class-incremental learning. Each of these scenarios has its own set of challenges. To illustrate this, we provide a comprehensive empirical comparison of currently used continual learning strategies, by performing the Split MNIST and Split CIFAR-100 protocols according to each scenario. We demonstrate substantial differences between the three scenarios in terms of difficulty and in terms of the effectiveness of different strategies. The proposed categorization aims to structure the continual learning field, by forming a key foundation for clearly defining benchmark problems.

Discovering novel visual categories from a set of unlabeled images is a crucial and essential capability for intelligent vision systems since it enables them to automatically learn new concepts with no need for human-annotated supervision anymore. To tackle this problem, existing approaches first pretrain a neural network with a set of labeled images and then fine-tune the network to cluster unlabeled images into a few categorical groups. However, their unified feature representation hits a tradeoff bottleneck between feature preservation on labeled data and feature adaptation on unlabeled data. To circumvent this bottleneck, we propose a residual-tuning approach, which estimates a new residual feature from the pretrained network and adds it with a previous basic feature to compute the clustering objective together. Our disentangled representation approach facilitates adjusting visual representations for unlabeled images and overcoming forgetting old knowledge acquired from labeled images, with no need of replaying the labeled images again. In addition, residual-tuning is an efficient solution, adding few parameters and consuming modest training time. Our results on three common benchmarks show consistent and considerable gains over other state-of-the-art methods, and further reduce the performance gap to the fully supervised learning setup. Moreover, we explore two extended scenarios, including using fewer labeled classes and continually discovering more unlabeled sets, where the results further signify the advantages and effectiveness of our residual-tuning approach against previous approaches. Our code is available at https://github.com/liuyudut/ResTune. ispartof: IEEE Transactions on Neural Networks and Learning Systems vol:PP pages:1-15 ispartof: location:United States status: Published online

Training models continually to detect and classify objects, from new classes and new domains, remains an open problem. In this work, we conduct a thorough analysis of why and how object detection models forget catastrophically. We focus on distillation-based approaches in two-stage networks; the most-common strategy employed in contemporary continual object detection work.Distillation aims to transfer the knowledge of a model trained on previous tasks -- the teacher -- to a new model -- the student -- while it learns the new task. We show that this works well for the region proposal network, but that wrong, yet overly confident teacher predictions prevent student models from effective learning of the classification head. Our analysis provides a foundation that allows us to propose improvements for existing techniques by detecting incorrect teacher predictions, based on current ground-truth labels, and by employing an adaptive Huber loss as opposed to the mean squared error for the distillation loss in the classification heads. We evidence that our strategy works not only in a class incremental setting, but also in domain incremental settings, which constitute a realistic context, likely to be the setting of representative real-world problems., Comment: Accepted at BMVC '22

Spatially dynamic neural networks adjust network execution based on the input data, saving computations by skipping non-important image regions. Yet, GPU implementations fail to achieve speedups from these spatially dynamic execution patterns for most neural network architectures. This paper investigates hardware constraints preventing such speedup and proposes and compares novel processor architectures and dataflows enabling latency improvements due to the dynamic execution with minimal loss of utilization. The presented architectures flexibly support spatial execution of a broad range of networks. For the derived architectures, the spatial unrolling for each layer type is optimized and validated making use of the ZigZag design space exploration framework where appropriate. This allows to benchmark and compare the hardware architectures on NNs for classification and human pose estimation, increasing throughput up to $\times 1.9$ and $\times 2.3$ compared to their static executions, respectively. This is the same order of magnitude as other dynamic execution methods, while being complementary to those.

We describe a deep-geometric localizer that is able to estimate the full six degrees-of-freedom (DOF) global pose of the camera from a single image in a previously mapped environment. Our map is a topo-metric one, with discrete topological nodes whose 6DOF poses are known. Each topo-node in our map also comprises of a set of points, whose 2D features and 3D locations are stored as part of the mapping process. For the mapping phase, we utilize a stereo camera and a regular stereo visual SLAM pipeline. During the localization phase, we take a single camera image, localize it to a topological node using deep learning, and use a geometric algorithm (perspective-n-point (PnP)) on the matched 2D features (and their 3D positions in the topo map) to determine the full 6DOF globally consistent pose of the camera. Our method divorces the mapping and the localization algorithms and sensors (stereo and mono) and allows accurate 6DOF pose estimation in a previously mapped environment using a single camera. With results in simulated and real environments, our hybrid algorithm is particularly useful for autonomous vehicles (AVs) and shuttles that might repeatedly traverse the same route.

To preserve the quality of fresh fruit after harvest and to meet the year-round demand for high-quality fruit, pears are stored under a controlled atmosphere. However, due to preharvest events or suboptimal storage conditions, internal disorders might develop resulting in severe quality loss. Examples include internal browning and cavities, which are invisible externally. Here, X-ray radiography is investigated as a technique for internal quality inspection. The detection of defect fruit is approached as an anomaly detection (AD) problem, in which a model is constructed using nominal data and an anomaly score is used to identify defect fruit. In this work, multiple deep AD methods are shown to be effective to detect pears with internal cavity and browning disorders using Xray radiographs (mean area under the receiver operating characteristic curve (AUC) up to 0.962). The best performing methods were found on par with a state-of-the-art multisensor disorder detection method (mean AUC up to 0.966). By investigating AD performance in function of internal disorder severity, it was shown that defect fruit with a cavity volume percentage > 1.0% could be detected 100% accurate using inline X-ray imaging. For lower cavity area percentages, the accuracy depended on the internal browning severity. Additionally, the explainability of the deep AD methods, i.e., how well human interpretable insight can be provided from each method's predictions, were qualitatively evaluated using anomaly heatmaps, which provided useful insight in the execution of the deep learning algorithms.

Anticipating the motion of dynamic objects is critical for making intelligent decisions navigating through an environment while avoiding collisions. In this work, we propose a CNN model that estimates 3D motion of objects using sequences of monocular images. We show that we can train this model without using any manual annotations by using Iterative Closest Points (ICP) to align pointclouds of an object at different points in time. We compare our unsupervised approach to a model that was trained using ground truth supervision, on the KITTI tracking dataset. We further improve our model by training our model on a larger dataset, which would otherwise not be possible due to the lack of ground truth data. We also compare our approach with a 3D object detector that estimates motion using a simple tracking scheme.

In this work, we propose a novel approach that allows for the end-to-end learning of multi-instance point detection with inherent sub-pixel precision capabilities. To infer unambiguous localization estimates, our model relies on three components: the continuous prediction capabilities of offset-regression-based models, the finer-grained spatial learning ability of a novel continuous heatmap matching loss function introduced to that effect, and the prediction sparsity ability of count-based regularization. We demonstrate strong sub-pixel localization accuracy on single molecule localization microscopy and checkerboard corner detection, and improved sub-frame event detection performance in sport videos.

We present MIX'EM, a novel solution for unsupervised image classification. MIX'EM generates representations that by themselves are sufficient to drive a general-purpose clustering algorithm to deliver high-quality classification. This is accomplished by building a mixture of embeddings module into a contrastive visual representation learning framework in order to disentangle representations at the category level. It first generates a set of embedding and mixing coefficients from a given visual representation, and then combines them into a single embedding. We introduce three techniques to successfully train MIX'EM and avoid degenerate solutions; (i) diversify mixture components by maximizing entropy, (ii) minimize instance conditioned component entropy to enforce a clustered embedding space, and (iii) use an associative embedding loss to enforce semantic separability. By applying (i) and (ii), semantic categories emerge through the mixture coefficients, making it possible to apply (iii). Subsequently, we run K-means on the representations to acquire semantic classification. We conduct extensive experiments and analyses on STL10, CIFAR10, and CIFAR100-20 datasets, achieving state-of-the-art classification accuracy of 78\%, 82\%, and 44\%, respectively. To achieve robust and high accuracy, it is essential to use the mixture components to initialize K-means. Finally, we report competitive baselines (70\% on STL10) obtained by applying K-means to the "normalized" representations learned using the contrastive loss.

Computer vision has shown great accomplishments in a wide variety of classification, segmentation and object recognition tasks, but tends to encounter more difficulties when tasks require more contextual assessment. Measuring the engagement of students is an example of such a complex task, as it requires a strong interpretative component. This research describes a methodology to measure students’ engagement, taking both an individual (student-level) and a collective (classroom) approach. Results show that students’ individual behaviour, such as note taking or hand-raising, is challenging to recognise, and does not correlate with students’ self reported engagement. Interestingly, students’ collective behaviour can be quantified in a more generic way using measures for students’ symmetry, reaction times and eye-gaze intersections. Nonetheless, the evidence for a connection between these collective measures and engagement is rather weak. Although this study does not succeed in providing a proxy of students’ self-reported engagement, our approach sheds light on the needs for future research. More concretely, we suggest that not only the behavioural, but also the emotional and cognitive component of engagement should be captured. ispartof: Mathematics vol:9 issue:3 pages:1-20 status: Published online

Learning from non-stationary data streams and overcoming catastrophic forgetting still poses a serious challenge for machine learning research. Rather than aiming to improve state-of-the-art, in this work we provide insight into the limits and merits of rehearsal, one of continual learning's most established methods. We hypothesize that models trained sequentially with rehearsal tend to stay in the same low-loss region after a task has finished, but are at risk of overfitting on its sample memory, hence harming generalization. We provide both conceptual and strong empirical evidence on three benchmarks for both behaviors, bringing novel insights into the dynamics of rehearsal and continual learning in general. Finally, we interpret important continual learning works in the light of our findings, allowing for a deeper understanding of their successes., Preprint, code publicly available

Given a really low resolution input image of a face (say \(16\,{\times }\,16\) or \(8\,{\times }\,8\) pixels), the goal of this paper is to reconstruct a high-resolution version thereof. This, by itself, is an ill-posed problem, as the high-frequency information is missing in the low-resolution input and needs to be hallucinated, based on prior knowledge about the image content. Rather than relying on a generic face prior, in this paper we explore the use of a set of exemplars, i.e. other high-resolution images of the same person. These guide the neural network as we condition the output on them. Multiple exemplars work better than a single one. To combine the information from multiple exemplars effectively, we introduce a pixel-wise weight generation module. Besides standard face super-resolution, our method allows to perform subtle face editing simply by replacing the exemplars with another set with different facial features. A user study is conducted and shows the super-resolved images can hardly be distinguished from real images on the CelebA dataset. A qualitative comparison indicates our model outperforms methods proposed in the literature on the CelebA and WebFace datasets.

Active visual exploration aims to assist an agent with a limited field of view to understand its environment based on partial observations made by choosing the best viewing directions in the scene. Recent methods have tried to address this problem either by using reinforcement learning, which is difficult to train, or by uncertainty maps, which are task-specific and can only be implemented for dense prediction tasks. In this paper, we propose the Glimpse-Attend-and-Explore model which: (a) employs self-attention to guide the visual exploration instead of task-specific uncertainty maps; (b) can be used for both dense and sparse prediction tasks; and (c) uses a contrastive stream to further improve the representations learned. Unlike previous works, we show the application of our model on multiple tasks like reconstruction, segmentation and classification. Our model provides encouraging results while being less dependent on dataset bias in driving the exploration. We further perform an ablation study to investigate the features and attention learned by our model. Finally, we show that our self-attention module learns to attend different regions of the scene by minimizing the loss on the downstream task. Code: https://github.com/soroushseifi/glimpse-attend-explore.

We address the problem of multimodal spatial understanding by decoding a set of language-expressed spatial relations to a set of 2D spatial arrangements in a multi-object and multi-relationship setting. We frame the task as arranging a scene of clip-arts given a textual description. We propose a simple and effective model architecture Spatial-Reasoning Bert (SR-Bert), trained to decode text to 2D spatial arrangements in a non-autoregressive manner. SR-Bert can decode both explicit and implicit language to 2D spatial arrangements, generalizes to out-of-sample data to a reasonable extent and can generate complete abstract scenes if paired with a clip-arts predictor. Finally, we qualitatively evaluate our method with a user study, validating that our generated spatial arrangements align with human expectation.

Overcoming catastrophic forgetting in neural networks is a long-standing and core research objective for incremental learning. Notable studies have shown regularization strategies enable the network to remember previously acquired knowledge devoid of heavy forgetting. Since those regularization strategies are mostly associated with classifier outputs, we propose a MUlti-Classifier (MUC) incremental learning paradigm that integrates an ensemble of auxiliary classifiers to estimate more effective regularization constraints. Additionally, we extend two common methods, focusing on parameter and activation regularization, from the conventional single-classifier paradigm to MUC. Our classifier ensemble promotes regularizing network parameters or activations when moving to learn the next task. Under the setting of task-agnostic evaluation, our experimental results on CIFAR-100 and Tiny ImageNet incremental benchmarks show that our method outperforms other baselines. Specifically, MUC obtains 3%–5% accuracy boost and 4%–5% decline of forgetting ratio, compared with MAS and LwF. Our code is available at https://github.com/Liuy8/MUC.

Mixture models are well-established learning approaches that, in computer vision, have mostly been applied to inverse or ill-defined problems. However, they are general-purpose divide-and-conquer techniques, splitting the input space into relatively homogeneous subsets in a data-driven manner. Not only ill-defined but also well-defined complex problems should benefit from them. To this end, we devise a framework for spatial regression using mixture density networks. We realize the framework for object detection and human pose estimation. For both tasks, a mixture model yields higher accuracy and divides the input space into interpretable modes. For object detection, mixture components focus on object scale, with the distribution of components closely following that of ground truth the object scale. This practically alleviates the need for multi-scale testing, providing a superior speed-accuracy trade-off. For human pose estimation, a mixture model divides the data based on viewpoint and uncertainty -- namely, front and back views, with back view imposing higher uncertainty. We conduct experiments on the MS COCO dataset and do not face any mode collapse.

This work investigates the task of unsupervised model personalization, adapted to continually evolving, unlabeled local user images. We consider the practical scenario where a high capacity server interacts with a myriad of resource-limited edge devices, imposing strong requirements on scalability and local data privacy. We aim to address this challenge within the continual learning paradigm and provide a novel Dual User-Adaptation framework (DUA) to explore the problem. This framework flexibly disentangles user-adaptation into model personalization on the server and local data regularization on the user device, with desirable properties regarding scalability and privacy constraints. First, on the server, we introduce incremental learning of task-specific expert models, subsequently aggregated using a concealed unsupervised user prior. Aggregation avoids retraining, whereas the user prior conceals sensitive raw user data, and grants unsupervised adaptation. Second, local user-adaptation incorporates a domain adaptation point of view, adapting regularizing batch normalization parameters to the user data. We explore various empirical user configurations with different priors in categories and a tenfold of transforms for MIT Indoor Scene recognition, and classify numbers in a combined MNIST and SVHN setup. Extensive experiments yield promising results for data-driven local adaptation and elicit user priors for server adaptation to depend on the model rather than user data. Hence, although user-adaptation remains a challenging open problem, the DUA framework formalizes a principled foundation for personalizing both on server and user device, while maintaining privacy and scalability., CVPR 2020

The estimation of the orientation of an observed vehicle relative to an Autonomous Vehicle (AV) from monocular camera data is an important building block in estimating its 6 DoF pose. Current Deep Learning based solutions for placing a 3D bounding box around this observed vehicle are data hungry and do not generalize well. In this paper, we demonstrate the use of monocular visual odometry for the self-supervised fine-tuning of a model for orientation estimation pre-trained on a reference domain. Specifically, while transitioning from a virtual dataset (vKITTI) to nuScenes, we recover up to 70% of the performance of a fully supervised method. We subsequently demonstrate an optimization-based monocular 3D bounding box detector built on top of the self-supervised vehicle orientation estimator without the requirement of expensive labeled data. This allows 3D vehicle detection algorithms to be self-trained from large amounts of monocular camera data from existing commercial vehicle fleets., Comment: ICRA 2021 (presentation)

We propose an approach without any forgetting to continual learning for the task-aware regime, where at inference the task-label is known. By using ternary masks we can upgrade a model to new tasks, reusing knowledge from previous tasks while not forgetting anything about them. Using masks prevents both catastrophic forgetting and backward transfer. We argue -- and show experimentally -- that avoiding the former largely compensates for the lack of the latter, which is rarely observed in practice. In contrast to earlier works, our masks are applied to the features (activations) of each layer instead of the weights. This considerably reduces the number of mask parameters for each new task; with more than three orders of magnitude for most networks. The encoding of the ternary masks into two bits per feature creates very little overhead to the network, avoiding scalability issues. To allow already learned features to adapt to the current task without changing the behavior of these features for previous tasks, we introduce task-specific feature normalization. Extensive experiments on several finegrained datasets and ImageNet show that our method outperforms current state-of-the-art while reducing memory overhead in comparison to weight-based approaches., Comment: To appear in the IEEE Conference on Computer Vision and Pattern Recognition Workshop (CVPR-W) on Continual Learning in Computer Vision (CLVision) 2021

In this paper, we develop a method for grounding medical text into a physically meaningful and interpretable space corresponding to a human atlas. We build on text embedding architectures such as Bert and introduce a loss function that allows us to reason about the semantic and spatial relatedness of medical texts by learning a projection of the embedding into a 3D space representing the human body. We quantitatively and qualitatively demonstrate that our proposed method learns a context sensitive and spatially aware mapping, in both the inter-organ and intra-organ sense, using a large scale medical text dataset from the “Large-scale online biomedical semantic indexing” track of the 2020 BioASQ challenge. We extend our approach to a self-supervised setting, and find it to be competitive with a classification based method, and a fully supervised variant of approach.

We propose a method to reduce the computational cost and memory consumption of existing neural networks, by exploiting spatial redundancies in images. Our method dynamically splits the image into blocks and processes low-complexity regions at a lower resolution. Our novel BlockPad module, implemented in CUDA, replaces zero-padding in order to prevent the discontinuities at patch borders of which existing methods suffer, while keeping memory consumption under control. We demonstrate SegBlocks on Cityscapes semantic segmentation, where the number of floating point operations is reduced by 30% with only 0.2% loss in accuracy (mIoU), and an inference speedup of 50% is achieved with 0.7% decrease in mIoU.

The task of visual grounding requires locating the most relevant region or object in an image, given a natural language query. So far, progress on this task was mostly measured on curated datasets, which are not always representative of human spoken language. In this work, we deviate from recent, popular task settings and consider the problem under an autonomous vehicle scenario. In particular, we consider a situation where passengers can give free-form natural language commands to a vehicle which can be associated with an object in the street scene. To stimulate research on this topic, we have organized the Commands for Autonomous Vehicles (C4AV) challenge based on the recent Talk2Car dataset. This paper presents the results of the challenge. First, we compare the used benchmark against existing datasets for visual grounding. Second, we identify the aspects that render top-performing models successful, and relate them to existing state-of-the-art models for visual grounding, in addition to detecting potential failure cases by evaluating on carefully selected subsets. Finally, we discuss several possibilities for future work.

Training deep learning models on embedded devices is typically avoided since this requires more memory, computation and power over inference. In this work, we focus on lowering the amount of memory needed for storing all activations, which are required during the backward pass to compute the gradients. Instead, during the forward pass, static Synthetic Gradient Modules (SGMs) predict gradients for each layer. This allows training the model in a feed-forward manner without having to store all activations. We tested our method on a robot grasping scenario where a robot needs to learn to grasp new objects given only a single demonstration. By first training the SGMs in a meta-learning manner on a set of common objects, during fine-tuning, the SGMs provided the model with accurate gradients to successfully learn to grasp new objects. We have shown that our method has comparable results to using standard backpropagation.

In a dynamic environment, an agent with a limited field of view/resource cannot fully observe the scene before attempting to parse it. The deployment of common semantic segmentation architectures is not feasible in such settings. In this paper we propose a method to gradually segment a scene given a sequence of partial observations. The main idea is to refine an agent’s understanding of the environment by attending the areas it is most uncertain about. Our method includes a self-supervised attention mechanism and a specialized architecture to maintain and exploit spatial memory maps for filling-in the unseen areas in the environment. The agent can select and attend an area while relying on the cues coming from the visited areas to hallucinate the other parts. We reach a mean pixel-wise accuracy of \(78.1\%\), \(80.9\%\) and \(76.5\%\) on CityScapes, CamVid, and Kitti datasets by processing only \(18\%\) of the image pixels (10 retina-like glimpses). We perform an ablation study on the number of glimpses, input image size and effectiveness of retina-like glimpses. We compare our method to several baselines and show that the optimal results are achieved by having access to a very low resolution view of the scene at the first timestep.

We propose a novel weakly supervised framework that jointly tackles entity analysis tasks in vision and language. Given a video with subtitles, we jointly address the questions: a) What do the textual entity mentions refer to? and b) What/ who are in the video key frames? We use a Markov Random Field (MRF) to encode the dependencies within and across the two modalities. This MRF model incorporates beliefs using independent methods for the textual and visual entities. These beliefs are propagated across the modalities to jointly derive the entity labels. We apply the framework to a challenging dataset of wildlife documentaries with subtitles and show that this integrated modelling yields significantly better performance over text-based and vision-based approaches. We show that textual mentions that cannot be resolved using text-only methods are resolved correctly using our method. The approaches described here bring us closer to automated multimedia indexing. Nurani Venkitasubramanian A., Tuytelaars T., Moens M.-F., ''Entity linking across vision and language'', Multimedia tools and applications, vol. 76, no. 21, pp. 22599-22622, 24 pp., November 2017. ispartof: Multimedia Tools and Applications vol:76 issue:21 pages:22599-22622 status: published

Telecommunication standards are documents that contain consolidated knowledge about communication systems and implementation best practices. They are created based on long consensus processes in order to meet practical constraints. This article describes how the DVB-S2 standard is used in the electrical engineering curricula at KU Leuven within a design and implementation course called EAGLE. The goal of the course is to teach third-year Bachelor students how to apply the abstract knowledge gathered in the theoretical courses (control theory, communication theory, software, and hardware) to the design of a complex and real engineering project. Based on their progress during the academic year, we illustrate how standards can be used to teach a broad skill set and act as an accelerator to abstract system implementation complexity. We give an overview of the objectives of the EAGLE project and the relevance of the DVB-S2 standard, the design and implementation task that the students need to accomplish, and how they finally learn to master the complexity of a real communication standard, VHDL coding, and system verification and integration.

We propose a function-based temporal pooling method that captures the latent structure of the video sequence data - e.g. how frame-level features evolve over time in a video. We show how the parameters of a function that has been fit to the video data can serve as a robust new video representation. As a specific example, we learn a pooling function via ranking machines. By learning to rank the frame-level features of a video in chronological order, we obtain a new representation that captures the video-wide temporal dynamics of a video, suitable for action recognition. Other than ranking functions, we explore different parametric models that could also explain the temporal changes in videos. The proposed functional pooling methods, and rank pooling in particular, is easy to interpret and implement, fast to compute and effective in recognizing a wide variety of actions. We evaluate our method on various benchmarks for generic action, fine-grained action and gesture recognition. Results show that rank pooling brings an absolute improvement of 7-10 average pooling baseline. At the same time, rank pooling is compatible with and complementary to several appearance and local motion based methods and features, such as improved trajectories and deep learning features., IEEE Transactions on Pattern Analysis and Machine Intelligence

In this work we propose an unsupervised training method that finetunes a single image depth estimation CNN towards a new environment. The network, which has been pretrained on stereo data, only requires monocular input for finetuning. Unlike other unsupervised methods, it produces depth estimations with absolute scale – a feature that is essential for most practical applications, yet has mostly been overlooked in the literature. First, we show how our method allows adapting a network trained on one dataset (Cityscapes) to another (KITTI). Next, by splitting KITTI in subsets, we show the sensitivity of pretrained models to a domain shift. We then demonstrate that, by finetuning the model using our method, it is possible to improve the performance on the target subset, without using stereo or any form of groundtruth depth and with preservation of the correct absolute scale.

This paper focuses on the problem of unsupervised image-to-image translation. More specifically, we aim at finding a translation network such that objects and shapes that only appear in the source domain are translated to objects and shapes only appearing in the target domain, while style color features present in the source domain remain the same. To achieve this, we use a domain-specific variational autoencoder and represent each image in its latent space representation. In a second step, we learn a translation between latent spaces of different domains using generative adversarial networks. We evaluate this framework on multiple datasets and verify the effect of multiple perceptual losses. Experiments on the MNIST and SVHN datasets show the effectiveness of the proposed translation method.

In this paper, a novel benchmark is introduced for evaluating local image descriptors. We demonstrate limitations of the commonly used datasets and evaluation protocols, that lead to ambiguities and contradictory results in the literature. Furthermore, these benchmarks are nearly saturated due to the recent improvements in local descriptors obtained by learning from large annotated datasets. To address these issues, we introduce a new large dataset suitable for training and testing modern descriptors, together with strictly defined evaluation protocols in several tasks such as matching, retrieval and verification. This allows for more realistic, thus more reliable comparisons in different application scenarios. We evaluate the performance of several state-of-the-art descriptors and analyse their properties. We show that a simple normalisation of traditional hand-crafted descriptors is able to boost their performance to the level of deep learning based descriptors once realistic benchmarks are considered. Additionally we specify a protocol for learning and evaluating using cross validation. We show that when training state-of-the-art descriptors on this dataset, the traditional verification task is almost entirely saturated.

In this paper, a novel benchmark is introduced for evaluating local image descriptors. We demonstrate limitations of the commonly used datasets and evaluation protocols, that lead to ambiguities and contradictory results in the literature. Furthermore, these benchmarks are nearly saturated due to the recent improvements in local descriptors obtained by learning from large annotated datasets. To address these issues, we introduce a new large dataset suitable for training and testing modern descriptors, together with strictly defined evaluation protocols in several tasks such as matching, retrieval and verification. This allows for more realistic, thus more reliable comparisons in different application scenarios. We evaluate the performance of several state-of-the-art descriptors and analyse their properties. We show that a simple normalisation of traditional hand-crafted descriptors is able to boost their performance to the level of deep learning based descriptors once realistic benchmarks are considered. Additionally we specify a protocol for learning and evaluating using cross validation. We show that when training state-of-the-art descriptors on this dataset, the traditional verification task is almost entirely saturated. ispartof: Ieee Transactions On Pattern Analysis And Machine Intelligence vol:42 issue:11 ispartof: location:United States status: Published online

So far life-long learning (LLL) has been studied in relatively small-scale and relatively artificial setups. Here, we introduce a new large-scale alternative. What makes the proposed setup more natural and closer to human-like visual systems is threefold: First, we focus on concepts (or facts, as we call them) of varying complexity, ranging from single objects to more complex structures such as objects performing actions, and objects interacting with other objects. Second, as in real-world settings, our setup has a long-tail distribution, an aspect which has mostly been ignored in the LLL context. Third, facts across tasks may share structure (e.g., \(\langle \)person, riding, wave\(\rangle \) and \(\langle \)dog, riding, wave\(\rangle \)). Facts can also be semantically related (e.g., “liger” relates to seen categories like “tiger” and “lion”). Given the large number of possible facts, a LLL setup seems a natural choice. To avoid model size growing over time and to optimally exploit the semantic relations and structure, we combine it with a visual semantic embedding instead of discrete class labels. We adapt existing datasets with the properties mentioned above into new benchmarks, by dividing them semantically or randomly into disjoint tasks. This leads to two large-scale benchmarks with 906,232 images and 165,150 unique facts, on which we evaluate and analyze state-of-the-art LLL methods.

In this paper we present an overview of the contributed work presented at the UAVision2018 ECCV workshop. This workshop focused on real-time image processing on-board of Unmanned Aerial Vehicles (UAVs). For such applications the computational complexity of state-of-the-art computer vision algorithms often conflicts with the need for real-time operation and the extreme resource limitations of the hardware. Apart from a summary of the accepted workshop papers, this work also aims to identify common challenges and concerns which were addressed by multiple authors during the workshop, and their proposed solutions.

Convolutional neural networks (CNNs) and transfer learning have recently been used for 6 degrees of freedom (6-DoF) camera pose estimation. While they do not reach the same accuracy as visual SLAM-based approaches and are restricted to a specific environment, they excel in robustness and can be applied even to a single image. In this paper, we study PoseNet [1] and investigate modifications based on datasets' characteristics to improve the accuracy of the pose estimates. In particular, we emphasize the importance of field-of-view over image resolution; we present a data augmentation scheme to reduce overfitting; we study the effect of Long-Short-Term-Memory (LSTM) cells. Lastly, we combine these modifications and improve PoseNet's performance for monocular CNN based camera pose regression., Comment: Accepted at Deep Learning for Visual SLAM workshop at ICCV 2019

Several new algorithms for camera-based fall detection have been proposed, with the aim of reliably alerting caregivers about older persons' falls at home. These algorithms have been evaluated almost exclusively using brief segments of video data captured in artificial environments under optimal conditions and with falls simulated by actors. By contrast, we collected real-life video data recorded over several months at seven older persons' residences. Here, we report on our fall-detection algorithm based on the state-of-the-art, and we present an analysis of the real-life video data. The performance of our detection algorithm was compared with the performance of three previously reported algorithms that used a publicly available simulation data set. All four algorithms produced similar results when using the simulated data. However, the performance of our algorithm degraded drastically when evaluating falls in the real-life data. The false alarm rate was especially high, showing that some challenges still need to be met to make the system sufficiently robust to deploy in real-world situations. We conclude that using more realistic data sets that include longer video recordings and a broad range of activities are essential to reveal weaknesses in fall-detection algorithms.

Humans can learn in a continuous manner. Old rarely utilized knowledge can be overwritten by new incoming information while important, frequently used knowledge is prevented from being erased. In artificial learning systems, lifelong learning so far has focused mainly on accumulating knowledge over tasks and overcoming catastrophic forgetting. In this paper, we argue that, given the limited model capacity and the unlimited new information to be learned, knowledge has to be preserved or erased selectively. Inspired by neuroplasticity, we propose a novel approach for lifelong learning, coined Memory Aware Synapses (MAS). It computes the importance of the parameters of a neural network in an unsupervised and online manner. Given a new sample which is fed to the network, MAS accumulates an importance measure for each parameter of the network, based on how sensitive the predicted output function is to a change in this parameter. When learning a new task, changes to important parameters can then be penalized, effectively preventing important knowledge related to previous tasks from being overwritten. Further, we show an interesting connection between a local version of our method and Hebb's rule,which is a model for the learning process in the brain. We test our method on a sequence of object recognition tasks and on the challenging problem of learning an embedding for predicting $$ triplets. We show state-of-the-art performance and, for the first time, the ability to adapt the importance of the parameters based on unlabeled data towards what the network needs (not) to forget, which may vary depending on test conditions., Comment: ECCV 2018

Training deep neural network policies end-to-end for real-world applications so far requires big demonstration datasets in the real world or big sets consisting of a large variety of realistic and closely related 3D CAD models. These real or virtual data should, moreover, have very similar characteristics to the conditions expected at test time. These stringent requirements and the time consuming data collection processes that they entail, are currently the most important impediment that keeps deep reinforcement learning from being deployed in real-world applications. Therefore, in this work we advocate an alternative approach, where instead of avoiding any domain shift by carefully selecting the training data, the goal is to learn a policy that can cope with it. To this end, we propose the DoShiCo challenge: to train a model in very basic synthetic environments, far from realistic, in a way that it can be applied in more realistic environments as well as take the control decisions on real-world data. In particular, we focus on the task of collision avoidance for drones. We created a set of simulated environments that can be used as benchmark and implemented a baseline method, exploiting depth prediction as an auxiliary task to help overcome the domain shift. Even though the policy is trained in very basic environments, it can learn to fly without collisions in a very different realistic simulated environment. Of course several benchmarks for reinforcement learning already exist - but they never include a large domain shift. On the other hand, several benchmarks in computer vision focus on the domain shift, but they take the form of a static datasets instead of simulated environments. In this work we claim that it is crucial to take the two challenges together in one benchmark., Comment: Published at SIMPAR 2018. Please visit the paper webpage for more information, a movie and code for reproducing results: https://kkelchte.github.io/doshico

© 2018 IEEE. The promise of self-driving cars promotes several advantages, e.g. they have the ability to outperform human drivers while being safer. Here we take a deeper look into some aspects from algorithms aimed at making this promise a reality. More specifically, we analyze an end-to-end neural network to predict a car's steering actions on a highway based on images taken from a single car-mounted camera. We focus our analysis on several aspects which could have a significant impact on the performance of the system. These aspects are: the input data format, the temporal dependencies between consecutive inputs, and the origin of the data. We show that, for the task at hand, regression networks outperform their classifier counterparts. In addition, there seems to be a small difference between networks that use coloured images and ones that use grayscale images as input. For the second aspect, by feeding the network three concatenated images, we get a significant decrease of 30% in mean squared error. For the third aspect, by using simulation data we are able to train networks that have a performance comparable to networks trained on real-life datasets. We also qualitatively demonstrate that the standard metrics that are used to evaluate networks do not necessarily accurately reflect a system's driving behaviour. We show that a promising confusion matrix may result in poor driving behaviour while a very ill-looking confusion matrix may result in good driving behaviour. Heylen J., Iven S., De Brabandere B., Oramas Mogrovejo J.A., Van Gool L., Tuytelaars T., ''From pixels to actions: Learning to drive a car with deep neural networks'', IEEE winter conference on applications of computer vision - WACV 2018, March 12-14, 2018, Lake Tahoe, Nevada, USA (accepted). ispartof: pages:606-615 ispartof: IEEE Winter Conference on Applications in Computer Vision vol:2018-January pages:606-615 ispartof: IEEE winter conference on applications of computer vision - WACV 2018 location:Lake Tahoe, Nevada, USA date:12 Mar - 14 Mar 2018 status: published

Images have a prominent role in the communication of news on the Web. We propose a novel method for image classification with subject categories when limited annotated images are available for training the classifier. A neural network based encoder learns image representations from paired news images and their texts. Once trained, this encoder transforms any image to a text-enriched representation of the image, which is then used as input for the classifier that categorizes an image according to its subject category. We have trained classifiers with different amounts of annotated images and found that the image classifier that uses the text-enriched image representations outperforms a baseline model that only uses image features especially in cases with limited training examples.

Methods proposed in the literature towards continual deep learning typically operate in a task-based sequential learning setup. A sequence of tasks is learned, one at a time, with all data of current task available but not of previous or future tasks. Task boundaries and identities are known at all times. This setup, however, is rarely encountered in practical applications. Therefore we investigate how to transform continual learning to an online setup. We develop a system that keeps on learning over time in a streaming fashion, with data distributions gradually changing and without the notion of separate tasks. To this end, we build on the work on Memory Aware Synapses, and show how this method can be made online by providing a protocol to decide i) when to update the importance weights, ii) which data to use to update them, and iii) how to accumulate the importance weights at each update step. Experimental results show the validity of the approach in the context of two applications: (self-)supervised learning of a face recognition model by watching soap series and learning a robot to avoid collisions., Comment: Accepted as a conference paper in CVPR 2019. Rahaf Aljundi and Klaas Kelchtermans have contributed equally to this work

In this paper, we demonstrate a system that automates the process of recording video lectures in classrooms. Through special hard-ware (lecturer and audience facing cameras and microphone arrays), we record multiple points of view of the lecture. Person detection and tracking, along with recognition of different human actions are used to digitally zoom in on the lecturer, and alternate focus between the lecturer and the slides or the blackboard. Audio sound source localization, along with face detection and tracking, is used to detect questions from the audience, to digitally zoom in on the member of the audience asking the question and to improve the quality of the sound recording. Finally, an automatic video editing system is used to naturally switch between the different video streams and to compose a compelling end product. We demonstrate the working system in two classrooms, over two 2-hour lectures, given by two lecturers. Hulens D., Aerts B., Chakravarty P., Diba A., Goedemé T., Roussel T., Zegers J., Tuytelaars T., Van Eycken L., Van Gool L., Van hamme H., Vennekens J., ''The CAMETRON lecture recording system: High quality video recording and editing with minimal human supervision'', Lecture notes in computer science, vol. 10704, pp. 518-530 (24th international conference on multiMedia modeling - MMM 2018, February 5-7, 2018, Bangkok, Thailand). ispartof: pages:518-530 ispartof: MultiMedia Modeling vol:10704 pages:518-530 ispartof: International conference on multimedia modeling - MMM 2018 location:Bangkok, Thailand date:5 Feb - 7 Feb 2018 status: published

This paper demonstrates the use of a single forward facing camera for obstacle avoidance on a quadrotor. We train a CNN for estimating depth from a single image. The depth map is then fed to a behaviour arbitration based control algorithm that steers the quadrotor away from obstacles. We conduct experiments with simulated and real drones in a variety of environments.

Subspace-based domain adaptation methods have been very successful in the context of image recognition. In this chapter, we discuss methods using Subspace Alignment (SA). They are based on a mapping function which aligns the source subspace with the target one, so as to obtain a domain invariant feature space. The solution of the corresponding optimization problem can be obtained in closed form, leading to a simple to implement and fast algorithm. The only hyperparameter involved corresponds to the dimension of the subspaces. We give two methods, SA and SA-MLE, for setting this variable. SA is a purely linear method. As a nonlinear extension, Landmarks-based Kernelized Subspace Alignment (LSSA) first projects the data nonlinearly based on a set of landmarks, which have been selected so as to reduce the discrepancy between the domains.

We introduce an approach for sketch classification based on Fisher vectors that significantly outperforms existing techniques. For the TU-Berlin sketch benchmark [Eitz et al. 2012a], our recognition rate is close to human performance on the same task. Motivated by these results, we propose a different benchmark for the evaluation of sketch classification algorithms. Our key idea is that the relevant aspect when recognizing a sketch is not the intention of the person who made the drawing, but the information that was effectively expressed. We modify the original benchmark to capture this concept more precisely and, as such, to provide a more adequate tool for the evaluation of sketch classification techniques. Finally, we perform a classification-driven analysis which is able to recover semantic aspects of the individual sketches, such as the quality of the drawing and the importance of each part of the sketch for the recognition.

International audience; Mid-level or semi-local features learnt using class-level information are potentially more distinctive than the traditional low-level local features constructed in a purely bottom-up fashion. At the same time they preserve some of the robustness properties with respect to occlusions and image clutter. In this paper we propose a new and effective scheme for extracting mid-level features for image classification, based on relevant pattern mining. In par- ticular, we mine relevant patterns of local compositions of densely sampled low-level features. We refer to the new set of obtained patterns as Frequent Local Histograms or FLHs. During this process, we pay special attention to keeping all the local histogram information and to selecting the most relevant reduced set of FLH patterns for classification. The careful choice of the visual primitives and an extension to exploit both local and global spatial information allow us to build powerful bag-of-FLH-based image representations. We show that these bag-of-FLHs are more discriminative than traditional bag-of-words and yield state-of-the-art results on various image classification benchmarks, including Pascal VOC.