Search

Understanding the loss surface of neural networks is essential for the design of models with predictable performance and their success in applications. Experimental results suggest that sufficiently deep and wide neural networks are not negatively impacted by suboptimal local minima. Despite recent progress, the reason for this outcome is not fully understood. Could deep networks have very few, if at all, suboptimal local optima or could all of them be equally good We provide a construction to show that suboptimal local minima (i.e., non-global ones), even though degenerate, exist for fully connected neural networks with sigmoid activation functions. The local minima obtained by our construction belong to a connected set of local solutions that can be escaped from via a non-increasing path on the loss curve. For extremely wide neural networks of decreasing width after the wide layer, we prove that every suboptimal local minimum belongs to such a connected set. This provides a partial explanation for the successful application of deep neural networks. In addition, we also characterize under what conditions the same construction leads to saddle points instead of local minima for deep neural networks. [ABSTRACT FROM AUTHOR]

Systems based on bag-of-words models from image features collected at maxima of sparse interest point operators have been used successfully for both computer visual object and action recognition tasks. While the sparse, interest-point based approach to recognition is not inconsistent with visual processing in biological systems that operate in ‘saccade and fixate’ regimes, the methodology and emphasis in the human and the computer vision communities remains sharply distinct. Here, we make three contributions aiming to bridge this gap. First, we complement existing state-of-the art large scale dynamic computer vision annotated datasets like Hollywood-2  [1] and UCF Sports [2] with human eye movements collected under the ecological constraints of visual action and scene context recognition tasks. To our knowledge these are the first large human eye tracking datasets to be collected and made publicly available for video, vision.imar.ro/eyetracking (497,107 frames, each viewed by 19 subjects), unique in terms of their (a) large scale and computer vision relevance, (b) dynamic, video stimuli, (c) task control, as well as free-viewing. Second, we introduce novel dynamic consistency and alignment measures, which underline the remarkable stability of patterns of visual search among subjects. Third, we leverage the significant amount of collected data in order to pursue studies and build automatic, end-to-end trainable computer vision systems based on human eye movements. Our studies not only shed light on the differences between computer vision spatio-temporal interest point image sampling strategies and the human fixations, as well as their impact for visual recognition performance, but also demonstrate that human fixations can be accurately predicted, and when used in an end-to-end automatic system, leveraging some of the advanced computer vision practice, can lead to state of the art results. [ABSTRACT FROM PUBLISHER]

Skeletonization algorithms typically decompose an object's silhouette into a set of symmetric parts, offering a powerful representation for shape categorization. However, having access to an object's silhouette assumes correct figure-ground segmentation, leading to a disconnect with the mainstream categorization community, which attempts to recognize objects from cluttered images. In this paper, we present a novel approach to recovering and grouping the symmetric parts of an object from a cluttered scene. We begin by using a multiresolution superpixel segmentation to generate medial point hypotheses, and use a learned affinity function to perceptually group nearby medial points likely to belong to the same medial branch. In the next stage, we learn higher granularity affinity functions to group the resulting medial branches likely to belong to the same object. The resulting framework yields a skeletal approximation that is free of many of the instabilities that occur with traditional skeletons. More importantly, it does not require a closed contour, enabling the application of skeleton-based categorization systems to more realistic imagery. [ABSTRACT FROM AUTHOR]

Detecting independent objects in images and videos is an important perceptual grouping problem. One common perceptual grouping cue that can facilitate this objective is the cue of contour closure, reflecting the spatial coherence of objects in the world and their projections as closed boundaries separating figure from background. Detecting contour closure in images consists of finding a cycle of disconnected contour fragments that separates an object from its background. Searching the entire space of possible groupings is intractable, and previous approaches have adopted powerful perceptual grouping heuristics, such as proximity and co-curvilinearity, to constrain the search. We introduce a new formulation of the problem, by transforming the problem of finding cycles of contour fragments to finding subsets of superpixels whose collective boundary has strong edge support (few gaps) in the image. Our cost function, a ratio of a boundary gap measure to area, promotes spatially coherent sets of superpixels. Moreover, its properties support a global optimization procedure based on parametric maxflow. Extending closure detection to videos, we introduce the concept of spatiotemporal closure. Analogous to image closure, we formulate our spatiotemporal closure cost over a graph of spatiotemporal superpixels. Our cost function is a ratio of motion and appearance discontinuity measures on the boundary of the selection to an internal homogeneity measure of the selected spatiotemporal volume. The resulting approach automatically recovers coherent components in images and videos, corresponding to objects, object parts, and objects with surrounding context, providing a good set of multiscale hypotheses for high-level scene analysis. We evaluate both our image and video closure frameworks by comparing them to other closure detection approaches, and find that they yield improved performance. [ABSTRACT FROM AUTHOR]

Abstract: One of the main shortcomings of Markov chain Monte Carlo samplers is their inability to mix between modes of the target distribution. In this paper we show that advance knowledge of the location of these modes can be incorporated into the MCMC sampler by introducing mode-hopping moves that satisfy detailed balance. The proposed sampling algorithm explores local mode structure through local MCMC moves (e.g. diffusion or Hybrid Monte Carlo) but in addition also represents the relative strengths of the different modes correctly using a set of global moves. This ‘mode-hopping’ MCMC sampler can be viewed as a generalization of the darting method . We illustrate the method on learning Markov random fields and evaluate it against the spherical darting algorithm on a ‘real world’ vision application of inferring 3D human body pose distributions from 2D image information. [Copyright &y& Elsevier]

We describe twin Gaussian processes (TGP), a generic structured prediction method that uses Gaussian process (GP) priors on both covariates and responses, both multivariate, and estimates outputs by minimizing the Kullback-Leibler divergence between two GP modeled as normal distributions over finite index sets of training and testing examples, emphasizing the goal that similar inputs should produce similar percepts and this should hold, on average, between their marginal distributions. TGP captures not only the interdependencies between covariates, as in a typical GP, but also those between responses, so correlations among both inputs and outputs are accounted for. TGP is exemplified, with promising results, for the reconstruction of 3d human poses from monocular and multicamera video sequences in the recently introduced HumanEva benchmark, where we achieve 5 cm error on average per 3d marker for models trained jointly, using data from multiple people and multiple activities. The method is fast and automatic: it requires no hand-crafting of the initial pose, camera calibration parameters, or the availability of a 3d body model associated with human subjects used for training or testing. [ABSTRACT FROM AUTHOR]

We introduce BM³E, a Conditional Bayesian Mixture of Experts Markov Model, that achieves consistent probabilistic estimates for discriminative visual tracking. The model applies to problems of temporal and uncertain inference and represents the unexplored bottom-up counterpart of pervasive generative models estimated with Kalman filtering or particle filtering. Instead of inverting a nonlinear generative observation model at runtime, we learn to cooperatively predict complex state distributions directly from descriptors that encode image observations (typically, bag-of-feature global image histograms or descriptors computed over regular spatial grids). These are integrated in a conditional graphical model in order to enforce temporal smoothness constraints and allow a principled management of uncertainty. The algorithms combine sparsity, mixture modeling, and nonlinear dimensionality reduction for efficient computation in high-dimensional continuous state spaces. The combined system automatically self-initializes and recovers from failure. The research has three contributions: 1) we establish the density propagation rules for discriminative inference in continuous, temporal chain models, 2) we propose flexible supervised and unsupervised algorithms to learn feed-forward, multivalued contextual mappings (multimodal state distributions) based on compact, conditional Bayesian mixture of experts models, and 3) we validate the framework empirically for the reconstruction of 3D human motion in monocular video sequences. Our tests on both real and motion-capture-based sequences show significant performance gains with respect to competing nearest neighbor, regression, and structured prediction methods. [ABSTRACT FROM AUTHOR]

Abstract: Sequential random sampling (‘Markov Chain Monte-Carlo’) is a popular strategy for many vision problems involving multi-modal distributions over high-dimensional parameter spaces. It applies both to importance sampling (where one wants to sample points according to their ‘importance’ for some calculation, but otherwise fairly) and to global-optimization (where one wants to find good minima, or at least good starting points for local minimization, regardless of fairness). Unfortunately, most sequential samplers are very prone to becoming trapped for long periods in unrepresentative local minima, which leads to biased or highly variable estimates. We present a general strategy for reducing MCMC trapping that generalizes Voter''s ‘hyperdynamic sampling’ from computational chemistry. The local gradient and curvature of the input distribution are used to construct an adaptive importance sampler that focuses samples on negative curvature regions that are likely to contain low cost ‘transition states’ (codimension-1 saddle points representing ‘mountain passes’ connecting adjacent cost basins). This substantially accelerates inter-basin transition rates while still preserving correct relative transition probabilities. Experimental tests on the difficult problem of 3D articulated human pose estimation from monocular images show significantly enhanced minimum exploration. [Copyright &y& Elsevier]

We present a model-based framework for incremental, adaptive object shape estimation and tracking in monocular image sequences. Parametric structure and motion estimation methods usually assume a fixed class of shape representation (splines, deformable superquadrics, etc.) that is initialized prior to tracking. Since the model shape coverage is fixed a priori, the incremental recovery of structure is decoupled from tracking, thereby limiting both processes in their scope and robustness. In this work, we describe a model-based framework that supports the automatic detection and integration of low-level geometric primitives (lines) incrementally. Such primitives are not explicitly captured in the initial model, but are moving consistently with its image motion. The consistency tests used to identify new structure are based on trinocular constraints between geometric primitives. The method allows not only an increase in the model scope, but also improves tracking accuracy by including the newly recovered features in its state estimation. The formulation is a step toward automatic model building, since it allows both weaker assumptions on the availability of a prior shape representation and on the number of features that would otherwise be necessary for entirely bottom-up reconstruction. We demonstrate the proposed approach on two separate image-based tracking domains, each involving complex 3D object structure and motion. [ABSTRACT FROM AUTHOR]

Becoming trapped in suboptimal local minima is a perennial problem when optimizing visual models, particularly in applications like monocular human body tracking where complicated parametric models are repeatedly fitted to ambiguous image measurements. We show that trapping can be significantly reduced by building ‘roadmaps’ of nearby minima linked by transition pathways—paths leading over low ‘mountain passes’ in the cost surface—found by locating the transition state (codimension-1 saddle point) at the top of the pass and then sliding downhill to the next minimum. We present two families of transition-state-finding algorithms based on local optimization. In eigenvector tracking, unconstrained Newton minimization is modified to climb uphill towards a transition state, while in hypersurface sweeping, a moving hypersurface is swept through the space and moving local minima within it are tracked using a constrained Newton method. These widely applicable numerical methods, which appear not to be known in vision and optimization, generalize methods from computational chemistry where finding transition states is critical for predicting reaction parameters. Experiments on the challenging problem of estimating 3D human pose from monocular images show that our algorithms find nearby transition states and minima very efficiently, but also underline the disturbingly large numbers of minima that can exist in this and similar model based vision problems. [ABSTRACT FROM AUTHOR]

We present a method for recovering three-dimensional (3D) human body motion from monocular video sequences based on a robust image matching metric, incorporation of joint limits and non-self-intersection constraints, and a new sample-and-refine search strategy guided by rescaled cost-function covariances. Monocular 3D body tracking is challenging: besides the difficulty of matching an imperfect, highly flexible, self-occluding model to cluttered image features, realistic body models have at least 30 joint parameters subject to highly nonlinear physical constraints, and at least a third of these degrees of freedom are nearly unobservable in any given monocular image. For image matching we use a carefully designed robust cost metric combining robust optical flow, edge energy; and motion boundaries. The nonlinearities and matching ambiguities make the parameter-space cost surface multimodal, ill-conditioned and highly nonlinear, so searching it is difficult. We discuss the limitations of CONDENSATION-like samplers, and describe a novel hybrid search algorithm that combines inflated-covariance-scaled sampling and robust continuous optimization subject to physical constraints and model priors. Our experiments on challenging monocular sequences show that robust cost modeling, joint and self-intersection constraints, and informed sampling are all essential for reliable monocular 3D motion estimation. [ABSTRACT FROM AUTHOR]

We present a novel framework to generate and rank plausible hypotheses for the spatial extent of objects in images using bottom-up computational processes and mid-level selection cues. The object hypotheses are represented as figure-ground segmentations, and are extracted automatically, without prior knowledge of the properties of individual object classes, by solving a sequence of Constrained Parametric Min-Cut problems (CPMC) on a regular image grid. In a subsequent step, we learn to rank the corresponding segments by training a continuous model to predict how likely they are to exhibit real-world regularities (expressed as putative overlap with ground truth) based on their mid-level region properties, then diversify the estimated overlap score using maximum marginal relevance measures. We show that this algorithm significantly outperforms the state of the art for low-level segmentation in the VOC 2009 and 2010 data sets. In our companion papers [1], [2], we show that the algorithm can be used, successfully, in a segmentation-based visual object category recognition pipeline. This architecture ranked first in the VOC2009 and VOC2010 image segmentation and labeling challenges. [ABSTRACT FROM PUBLISHER]

Human motion analysis in images and video, with its deeply inter-related 2D and 3D inference components, is a central computer vision problem. Yet, there are no studies that reveal how humans perceive other people in images and how accurate they are. In this paper we aim to unveil some of the processing-as well as the levels of accuracy-involved in the 3D perception of people from images by assessing the human performance. Moreover, we reveal the quantitative and qualitative differences between human and computer performance when presented with the same visual stimuli and show that metrics incorporating human perception can produce more meaningful results when integrated into automatic pose prediction algorithms. Our contributions are: (1) the construction of an experimental apparatus that relates perception and measurement, in particular the visual and kinematic performance with respect to 3D ground truth when the human subject is presented an image of a person in a given pose; (2) the creation of a dataset containing images, articulated 2D and 3D pose ground truth, as well as synchronized eye movement recordings of human subjects, shown a variety of human body configurations, both easy and difficult, as well as their 're-enacted' 3D poses; (3) quantitative analysis revealing the human performance in 3D pose re-enactment tasks, the degree of stability in the visual fixation patterns of human subjects, and the way it correlates with different poses; (4) extensive analysis on the differences between human re-enactments and poses produced by an automatic system when presented with the same visual stimuli; (5) an approach to learning perceptual metrics that, when integrated into visual sensing systems, produces more stable and meaningful results. [ABSTRACT FROM AUTHOR]

Boundary detection is a fundamental computer vision problem that is essential for a variety of tasks, such as contour and region segmentation, symmetry detection and object recognition and categorization. We propose a generalized formulation for boundary detection, with closed-form solution, applicable to the localization of different types of boundaries, such as object edges in natural images and occlusion boundaries from video. Our generalized boundary detection method (Gb) simultaneously combines low-level and mid-level image representations in a single eigenvalue problem and solves for the optimal continuous boundary orientation and strength. The closed-form solution to boundary detection enables our algorithm to achieve state-of-the-art results at a significantly lower computational cost than current methods. We also propose two complementary novel components that can seamlessly be combined with Gb: first, we introduce a soft-segmentation procedure that provides region input layers to our boundary detection algorithm for a significant improvement in accuracy, at negligible computational cost; second, we present an efficient method for contour grouping and reasoning, which when applied as a final post-processing stage, further increases the boundary detection performance. [ABSTRACT FROM AUTHOR]

We propose a layered statistical model for image segmentation and labeling obtained by combining independently extracted, possibly overlapping sets of figure-ground (FG) segmentations. The process of constructing consistent image segmentations, called tilings, is cast as optimization over sets of maximal cliques sampled from a graph connecting all non-overlapping figure-ground segment hypotheses. Potential functions over cliques combine unary, Gestalt-based figure qualities, and pairwise compatibilities among spatially neighboring segments, constrained by T-junctions and the boundary interface statistics of real scenes. Building on the segmentation layer, we further derive a joint image segmentation and labeling model (JSL) which, given a bag of FGs, constructs a joint probability distribution over both the compatible image interpretations ( tilings) composed from those segments, and over their labeling into categories. The process of drawing samples from the joint distribution can be interpreted as first sampling tilings, followed by sampling labelings conditioned on the choice of a particular tiling. We learn the segmentation and labeling parameters jointly, based on maximum likelihood with a novel estimation procedure we refer to as incremental saddle-point approximation. The partition function over tilings and labelings is increasingly more accurately approximated by including incorrect configurations that are rated as probable by candidate models during learning. State of the art results are reported on the Berkeley, Stanford and Pascal VOC datasets, where an improvement of 28 % was achieved for the segmentation task only (tiling), and an accuracy of 47.8 % was obtained on the test set of VOC12 for semantic labeling (JSL). [ABSTRACT FROM AUTHOR]

We present an approach to visual object-class segmentation and recognition based on a pipeline that combines multiple figure-ground hypotheses with large object spatial support, generated by bottom-up computational processes that do not exploit knowledge of specific categories, and sequential categorization based on continuous estimates of the spatial overlap between the image segment hypotheses and each putative class. We differ from existing approaches not only in our seemingly unreasonable assumption that good object-level segments can be obtained in a feed-forward fashion, but also in formulating recognition as a regression problem. Instead of focusing on a one-vs.-all winning margin that may not preserve the ordering of segment qualities inside the non-maximum (non-winning) set, our learning method produces a globally consistent ranking with close ties to segment quality, hence to the extent entire object or part hypotheses are likely to spatially overlap the ground truth. We demonstrate results beyond the current state of the art for image classification, object detection and semantic segmentation, in a number of challenging datasets including Caltech-101, ETHZ-Shape as well as PASCAL VOC 2009 and 2010. [ABSTRACT FROM AUTHOR]

Computer Vision has been revolutionized by deep learning, and it is not surprising that the overwhelming majority of our selected papers make extensive use of neural networks. The flow of ideas between Computer Vision and Deep Learning continues to be bidirectional, and one of the selected papers defines a new learning procedure that may prove to be useful beyond computer vision (Wu and He [11]). The idea for a special issue about architectures and theories for computer vision came from the ECCV conference held in 2018. [Extracted from the article]

The role of symmetry in computer vision has waxed and waned in importance during the evolution of the field from its earliest days. At first figuring prominently in support of bottom-up indexing, it fell out of favour as shape gave way to appearance and recognition gave way to detection. With a strong prior in the form of a target object, the role of the weaker priors offered by perceptual grouping was greatly diminished. However, as the field returns to the problem of recognition from a large database, the bottom-up recovery of the parts that make up the objects in a cluttered scene is critical for their recognition. The medial axis community has long exploited the ubiquitous regularity of symmetry as a basis for the decomposition of a closed contour into medial parts. However, today's recognition systems are faced with cluttered scenes and the assumption that a closed contour exists, i.e., that figure-ground segmentation has been solved, rendering much of the medial axis community's work inapplicable. In this article, we review a computational framework, previously reported in [1-3], that bridges the representation power of the medial axis and the need to recover and group an object's parts in a cluttered scene. Our framework is rooted in the idea that a maximally-inscribed disc, the building block of a medial axis, can be modelled as a compact superpixel in the image. We evaluate the method on images of cluttered scenes. [ABSTRACT FROM AUTHOR]

Semantic segmentation and object detection are nowadays dominated by methods operating on regions obtained as a result of a bottom-up grouping process (segmentation) but use feature extractors developed for recognition on fixed-form (e.g. rectangular) patches, with full images as a special case. This is most likely suboptimal. In this paper we focus on feature extraction and description over free-form regions and study the relationship with their fixed-form counterparts. Our main contributions are novel pooling techniques that capture the second-order statistics of local descriptors inside such free-form regions. We introduce second-order generalizations of average and max-pooling that together with appropriate non-linearities, derived from the mathematical structure of their embedding space, lead to state-of-the-art recognition performance in semantic segmentation experiments without any type of local feature coding. In contrast, we show that codebook-based local feature coding is more important when feature extraction is constrained to operate over regions that include both foreground and large portions of the background, as typical in image classification settings, whereas for high-accuracy localization setups, second-order pooling over free-form regions produces results superior to those of the winning systems in the contemporary semantic segmentation challenges, with models that are much faster in both training and testing. [ABSTRACT FROM PUBLISHER]

We introduce a new dataset, Human3.6M, of 3.6 Million accurate 3D Human poses, acquired by recording the performance of 5 female and 6 male subjects, under 4 different viewpoints, for training realistic human sensing systems and for evaluating the next generation of human pose estimation models and algorithms. Besides increasing the size of the datasets in the current state-of-the-art by several orders of magnitude, we also aim to complement such datasets with a diverse set of motions and poses encountered as part of typical human activities (taking photos, talking on the phone, posing, greeting, eating, etc.), with additional synchronized image, human motion capture, and time of flight (depth) data, and with accurate 3D body scans of all the subject actors involved. We also provide controlled mixed reality evaluation scenarios where 3D human models are animated using motion capture and inserted using correct 3D geometry, in complex real environments, viewed with moving cameras, and under occlusion. Finally, we provide a set of large-scale statistical models and detailed evaluation baselines for the dataset illustrating its diversity and the scope for improvement by future work in the research community. Our experiments show that our best large-scale model can leverage our full training set to obtain a 20% improvement in performance compared to a training set of the scale of the largest existing public dataset for this problem. Yet the potential for improvement by leveraging higher capacity, more complex models with our large dataset, is substantially vaster and should stimulate future research. The dataset together with code for the associated large-scale learning models, features, visualization tools, as well as the evaluation server, is available online at http://vision.imar.ro/human3.6m. [ABSTRACT FROM AUTHOR]