CVPR 2015 Papers

In video based face recognition, great success has been made by representing videos as linear subspaces, which typically lie in a special type of non-Euclidean space known as Grassmann manifold. To leverage the kernel-based methods developed for Euclidean space, several recent methods have been proposed to embed the Grassmann manifold into a high dimensional Hilbert space by exploiting the well established Project Metric, which can approximate the Riemannian geometry of Grassmann manifold. Nevertheless, they inevitably introduce the drawbacks from traditional kernel-based methods such as implicit map and high computational cost to the Grassmann manifold. To overcome such limitations, we propose a novel method to learn the Projection Metric directly on Grassmann manifold rather than in Hilbert space. From the perspective of manifold learning, our method can be regarded as performing a geometry-aware dimensionality reduction from the original Grassmann manifold to a lower-dimensional, more discriminative Grassmann manifold where more favorable classification can be achieved. Experiments on several real-world video face datasets demonstrate that the proposed method yields competitive performance compared with the state-of-the-art algorithms.

Cascaded regression approaches have been recently shown to achieve state-of-the-art performance for many computer vision tasks. Beyond its connection to boosting, cascaded regression has been interpreted as a learning-based approach to iterative optimization methods like the Newton's method. However, in prior work, the connection to optimization theory is limited only in learning a mapping from image features to problem parameters. In this paper, we consider the problem of facial deformable model fitting using cascaded regression and make the following contributions: (a) We propose regression to learn a sequence of averaged Jacobian and Hessian matrices from data, and from them descent directions in a fashion inspired by Gauss-Newton optimization. (b) We show that the optimization problem in hand has structure and devise a learning strategy for a cascaded regression approach that takes the problem structure into account. By doing so, the proposed method learns and employs a sequence of averaged Jacobians and descent directions in a subspace orthogonal to the facial appearance variation; hence, we call it Project-Out Cascaded Regression (PO-CR). (c) Based on the principles of PO-CR, we built a face alignment system that produces remarkably accurate results on the challenging iBUG data set outperforming previously proposed systems by a large margin. Code for our system is available from http://www.cs.nott.ac.uk/~yzt/.

We propose a novel approach to computing the probabilities of presence of multiple and potentially occluding objects in a scene from a single depth map. To this end, we use a generative model that predicts the distribution of depth images that would be produced if the probabilities of presence were known and then to optimize them so that this distribution explains observed evidence as closely as possible. This allows us to exploit very effectively the available evidence and outperform state-of-the-art methods without requiring large amounts of data, or without using the RGB signal that modern RGB-D sensors also provide.

The next wave of micro and nano devices will create a world with trillions of small networked cameras. This will lead to increased concerns about privacy and security. Most privacy preserving algorithms for computer vision are applied after image/video data has been captured. We propose to use privacy preserving optics that filter or block sensitive information directly from the incident light-field before sensor measurements are made, adding a new layer of privacy. In addition to balancing the privacy and utility of the captured data, we address trade-offs unique to miniature vision sensors, such as achieving high-quality field-of-view and resolution within the constraints of mass and volume. Our privacy preserving optics enable applications such as depth sensing, full-body motion tracking, people counting, blob detection and privacy preserving face recognition. While we demonstrate applications on macro-scale devices (smartphones, webcams, etc.) our theory has impact for smaller devices.

Previous work on predicting the target of visual search from human fixations only considered closed-world settings in which training labels are available and predictions are performed for a known set of potential targets. In this work we go beyond the state of the art by studying search target prediction in an open-world setting in which we no longer assume that we have fixation data to train for the search targets. We present a dataset containing fixation data of 18 users searching for natural images from three image categories within synthesised image collages of about 80 images. In a closed-world baseline experiment we show that we can predict the correct target image out of a candidate set of five images. We then present a new problem formulation for search target prediction in the open-world setting that is based on learning compatibilities between fixations and potential targets.

We study the problem of predicting the future, though only in the probabilistic sense of estimating a future state of a time-varying probability distribution. This is not only an interesting academic problem, but solving this extrapolation problem also has many practical application, e.g. for training classifiers that have to operate under time-varying conditions. Our main contribution is a method for predicting the next step of the time-varying distribution from a given sequence of sample sets from earlier time steps. For this we rely on two recent machine learning techniques: embedding probability distributions into a reproducing kernel Hilbert space, and learning operators by vector-valued regression. We illustrate the working principles and the practical usefulness of our method by experiments on synthetic and real data. We also highlight an exemplary application: training a classifier in a domain adaptation setting without having access to examples from the test time distribution at training time.

It is believed that eye movements in free-viewing of natural scenes are directed by both bottom-up visual saliency and top-down visual factors. In this paper, we propose a novel computational framework to simultaneously learn these two types of visual features from raw image data using a multiresolution convolutional neural network (Mr-CNN) for predicting eye fixations. The Mr-CNN is directly trained from image regions centered on fixation and non-fixation locations over multiple resolutions, using raw image pixels as inputs and eye fixation attributes as labels. Diverse top-down visual features can be learned in higher layers. Meanwhile bottom-up visual saliency can also be inferred via combining information over multiple resolutions. Finally, optimal integration of bottom-up and top-down cues can be learned in the last logistic regression layer to predict eye fixations. The proposed approach achieves state-of-the-art results over four publically available benchmark datasets, demonstrating the superiority of our work.

Outliers pose a problem in all real structure from motion systems. Due to the use of automatic matching methods one has to expect that a (sometimes very large) portion of the detected correspondences can be incorrect. In this paper we propose a method that estimates the relative translation between two cameras and simultaneously maximizes the number of inlier correspondences. Traditionally, outlier removal tasks have been addressed using RANSAC approaches. However, these are random in nature and offer no guarantees of finding a good solution. If the amount of mismatches is large, the approach becomes costly because of the need to evaluate a large number of random samples. In contrast, our approach is based on the branch and bound methodology which guarantees that an optimal solution will be found. While most optimal methods trade speed for optimality, the proposed algorithm has competitive running times on problem sizes well beyond what is common in practice. Experiments on both real and synthetic data show that the method outperforms state-of-the-art alternatives, including RANSAC, in terms of solution quality. In addition, the approach is shown to be faster than RANSAC in settings with a large amount of outliers.

Estimating 3D human pose from 2D joint locations is central to the analysis of people in images and video. To address the fact that the problem is inherently ill posed, many methods impose a prior over human poses. Unfortunately these priors admit invalid poses because they do not model how joint-limits vary with pose. Here we make two key contributions. First, we collect a motion capture dataset that explores a wide range of human poses. From this we learn a pose-dependent model of joint limits that forms our prior. Both dataset and prior are available for research purposes. Second, we define a general parametrization of body pose and a new, multi-stage, method to estimate 3D pose from 2D joint locations using an over-complete dictionary of poses. Our method shows good generalization while avoiding impossible poses. We quantitatively compare our method with recent work and show state-of-the-art results on 2D to 3D pose estimation using the CMU mocap dataset. We also show superior results using manual annotations on real images and automatic detections on the Leeds sports pose dataset.

In this paper, we present a new feature representation for first-person videos. In first-person video understanding (e.g., activity recognition), it is very important to capture both entire scene dynamics (i.e., egomotion) and salient local motion observed in videos. We describe a representation framework based on time series pooling, which is designed to abstract short-term/long-term changes in feature descriptor elements. The idea is to keep track of how descriptor values are changing over time and summarize them to represent motion in the activity video. The framework is general, handling any types of per-frame feature descriptors including conventional motion descriptors like histogram of optical flows (HOF) as well as appearance descriptors from more recent convolutional neural networks (CNN). We experimentally confirm that our approach clearly outperforms previous feature representations including bag-of-visual-words and improved Fisher vector (IFV) when using identical underlying feature descriptors. We also confirm that our feature representation has superior performance to existing state-of-the-art features like local spatio-temporal features and Improved Trajectory Features (originally developed for 3rd-person videos) when handling first-person videos. Multiple first-person activity datasets were tested under various settings to confirm these findings.

Recent progress on probabilistic modeling and statistical learning, coupled with the availability of large training datasets, has led to remarkable progress in computer vision. Generative probabilistic models, or analysis-by-synthesis approaches, can capture rich scene structure but have been less widely applied than their discriminative counterparts, as they often require considerable problem-specific engineering in modeling and inference, and inference is typically seen as requiring slow, hypothesize-and-test Monte Carlo methods. Here we present Picture, a probabilistic programming language for scene understanding that allows researchers to express complex generative vision models, while automatically solving them using fast general-purpose inference machinery. Picture provides a stochastic scene language that can express generative models for arbitrary 2D/3D scenes, as well as a hierarchy of representation layers for comparing scene hypotheses with observed images by matching not simply pixels, but also more abstract features (e.g., contours, deep neural network activations). Inference can flexibly integrate advanced Monte Carlo strategies with fast bottom-up data-driven methods. Thus both representations and inference strategies can build directly on progress in discriminatively trained systems to make generative vision more robust and efficient. We use Picture to write programs for 3D face analysis, 3D human pose estimation, and 3D object reconstruction - each competitive with specially engineered baselines.

We tackle the problem of photometric stereo under near point lighting in this paper. Different from the conventional formulation of photometric stereo that assumes parallel lighting, photometric stereo under the near point lighting condition is a nonlinear problem as the local surface normals are coupled with its distance to the camera as well as the light sources. To solve this non-linear problem of PS with near point lighting, a local/global mesh deformation approach is developed in our work to determine the position and the orientation of a facet simultaneously, where each facet is corresponding to a pixel in the image captured by the camera. Unlike nonlinear optimization schemes, the mesh deformation in our approach is decoupled into an iteration of interlaced steps of local projection and global blending. Experimental results verify that our method can generate accurate estimation of surface shape under near point lighting in a few iterations. Besides, this approach is robust to errors on the positions of light sources and is easy to be implemented.

In this paper, we propose a novel uncalibrated photometric method for refining depth maps of multi-albedo objects obtained from consumer depth cameras like Kinect. Existing uncalibrated photometric methods either assume that the object has constant albedo or rely on segmenting images into constant albedo regions. The method of this paper does not require the constant albedo assumption and we believe it is the first work of its kind to handle objects with arbitrary albedo under uncalibrated illumination. We first robustly estimate a rank 3 approximation of the observed brightness matrix using an iterative reweighting method. Subsequently, we factorize this rank reduced brightness matrix into the corresponding lighting, albedo and surface normal components. The proposed factorization is shown to be convergent. We experimentally demonstrate the value of our approach by presenting highly accurate three-dimensional reconstructions of a wide variety of objects. Additionally, since any photometric method requires a radiometric calibration of the camera used, we also present a direct radiometric calibration technique for the infra-red camera of the structured-light stereo depth scanner. Unlike existing methods, this calibration technique does not depend on a known calibration object or on the properties of the scene illumination used.

Standard approaches to computing interpolated (in-between) frames in a video sequence require accurate pixel correspondences between images e.g. using optical flow. We present an efficient alternative by leveraging recent developments in phase-based methods that represent motion in the phase shift of individual pixels. This concept allows in-between images to be generated by simple per-pixel phase modification, without the need for any form of explicit correspondence estimation. Up until now, such methods have been limited in the range of motion that can be interpolated, which fundamentally restricts their usefulness. In order to reduce these limitations, we introduce a novel, bounded phase shift correction method that combines phase information across the levels of a multi-scale pyramid. Additionally, we propose extensions for phase-based image synthesis that yield smoother transitions between the interpolated images. Our approach avoids expensive global optimization typical of optical flow methods, and is both simple to implement and easy to parallelize. This allows us to interpolate frames at a fraction of the computational cost of traditional optical flow-based solutions, while achieving similar quality and in some cases even superior results. Our method fails gracefully in difficult interpolation settings, e.g., significant appearance changes, where flow-based methods often introduce serious visual artifacts. Due to its efficiency, our method is especially well suited for frame interpolation and retiming of high resolution, high frame rate video.

Person re-identification is an important technique towards automatic search of a person's presence in a surveillance video. Two fundamental problems are critical for person re-identification, feature representation and metric learning. An effective feature representation should be robust to illumination and viewpoint changes, and a discriminant metric should be learned to match various person images. In this paper, we propose an effective feature representation called Local Maximal Occurrence (LOMO), and a subspace and metric learning method called Cross-view Quadratic Discriminant Analysis (XQDA). The LOMO feature analyzes the horizontal occurrence of local features, and maximizes the occurrence to make a stable representation against viewpoint changes. Besides, to handle illumination variations, we apply the Retinex transform and a scale invariant texture operator. To learn a discriminant metric, we propose to learn a discriminant low dimensional subspace by cross-view quadratic discriminant analysis, and simultaneously, a QDA metric is learned on the derived subspace. We also present a practical computation method for XQDA, as well as its regularization. Experiments on four challenging person re-identification databases, VIPeR, QMUL GRID, CUHK Campus, and CUHK03, show that the proposed method improves the state-of-the-art rank-1 identification rates by 2.2%, 4.88%, 28.91%, and 31.55% on the four databases, respectively.

This paper formulates and presents a solution to a new problem called person count localization. Given a video of a crowded scene, our goal is to output for each frame a set of: 1) Detections optimally covering both isolated individuals and cluttered groups of people; and 2) Counts of people inside these detections. This problem is a middle-ground between frame-level person counting, which does not localize counts, and person detection aimed at perfectly localizing people with count-one detections. Our problem formulation is important for a wide range of domains, where people appear frequently under severe occlusion within a crowd. As these crowds are often visually distinct from the rest of the scene, they can be viewed as ``visual phrases'' whose spatially tight localization and count assignment could facilitate higher-level video understanding. For count localization, we specify a novel framework of iterative error-driven revisions of a flow graph derived from noisy input of people detections and foreground segmentation. Each iteration creates and solves an integer program for count localization based on iterative revisions of the flow graph. The graph revisions are based on detected violations of basic integrity constraints. They in turn trigger learned modifications to the graph aimed at reducing noise in input features. For evaluation, we introduce a new metric that measures both count precision and localization of our approach on American football and pedestrian videos.

Deep learning methods have achieved great success in pedestrian detection, owing to its ability to learn features from raw pixels. However, they can mainly capture middlelevel representations, such as pose of pedestrian, but confuses positive with hard negative samples (Fig.1 (a)), which have large ambiguity and can only be distinguished by highlevel representation. To address this ambiguity, this work jointly optimize pedestrian detection with semantic tasks, including pedestrian attributes (e.g. 'carrying backpack') and scene attributes (e.g. 'vehicle', 'tree', and 'horizontal'). Rather than expensively annotating scene attributes, we transfer attributes information from existing scene segmentation datasets to the pedestrian dataset, by proposing a novel deep model to learn high-level features from multiple tasks and multiple data sources. Since distinct tasks have distinct convergence rates and data from different datasets have different distributions, a multi-task objective function is carefully designed to coordinate tasks and reduce discrepancies among datasets. The importance coefficients of tasks and network parameters in this objective function can be iteratively estimated. Extensive evaluations show that the proposed approach outperforms the state-of-the-art on the challenging Caltech [10] and ETH [11] datasets where it reduces the miss rates of previous deep models by 17 and 5.5 percent, respectively.

Object segmentation is highly desirable for image understanding and editing. Current interactive tools require a great deal of user effort while automatic methods are usually limited to images of special object categories or with high color contrast. In this paper, we propose a data-driven algorithm that uses examples to break through these limits. As similar objects tend to share similar local shapes, we match query image patches with example images in multiscale to enable local shape transfer. The transferred local shape masks constitute a patch-level segmentation solution space and we thus develop a novel cascade algorithm, PatchCut, for coarse-to-fine object segmentation. In each stage of the cascade, local shape mask candidates are selected to refine the estimated segmentation of the previous stage iteratively with color models. Experimental results on various datasets (Weizmann Horse, Fashionista, Object Discovery and PASCAL) demonstrate the effectiveness and robustness of our algorithm.

We propose a method to recover the structure of a compound scene from multiple silhouettes. Structure is expressed as a collection of 3D primitives chosen from a pre-defined library, each with an associated pose. This has several advantages over a volume or mesh representation both for estimation and the utility of the recovered model. The main challenge in recovering such a model is the combinatorial number of possible arrangements of parts. We address this issue by exploiting the intrinsic structure and sparsity of the problem, and show that our method scales to scenes constructed from large libraries of parts.