CVPR 2015 Papers

This paper presents a hierarchical framework for detecting local and global anomalies via hierarchical feature representation and Gaussian process regression. While local anomaly is typically detected as a 3D pattern matching problem, we are more interested in global anomaly that involves multiple normal events interacting in an unusual manner such as car accident. To simultaneously detect local and global anomalies, we formulate the extraction of normal interactions from training video as the problem of efficiently finding the frequent geometric relations of the nearby sparse spatio-temporal interest points. A codebook of interaction templates is then constructed and modeled using Gaussian process regression. A novel inference method for computing the likelihood of an observed interaction is also proposed. As such, our model is robust to slight topological deformations and can handle the noise and data unbalance problems in the training data. Simulations show that our system outperforms the main state-of-the-art methods on this topic and achieves at least 80% detection rates based on three challenging datasets.

People preserve memories of events such as birthdays, weddings, or vacations by capturing photos, often depicting groups of people. Invariably, some individuals in the image are more important than others given the context of the event. This paper analyzes the concept of the importance of individuals in group photographs. We address two specific questions - Given an image, who are the most important individuals in it? Given multiple images of a person, which image depicts the person in the most important role? We introduce a measure of importance of people in images and investigate the correlation between importance and visual saliency. We find that not only can we automatically predict the importance of people from purely visual cues, incorporating this predicted importance results in significant improvement in applications such as im2text (generating sentences that describe images of groups of people).

Alignment of semantically meaningful visual patterns, such as object classes, is an important pre-processing step for a number of applications such as object detection and image categorization. Considering the expensive manpower spent on the annotation of supervised alignment methods, unsupervised alignment techniques are more favourable especially for large-scale problems. Fine adjustment can be effectively and efficiently achieved with the recent image congealing methods, but they require moderately good initialisation which is largely invalid in practice. It remains as an open problem how to align images of class examples with large view point changes. Feature-based methods can solve the problem to some degree, but require manual selection of a good seed image and omit the fact that two examples of semantical classes can be visually very different (e.g., Harley-Davison and Scooter``motorbikes''). In this work, we adopt the feature basedapproach,but to overcome the aforementioned drawbacks define visual similarity as an assignment problem which is solved by fast approximation and non-linear optimization.From pair-wise image similarities we construct an image graph which is used to step-wise align,``morph'', an image to another by graph traveling. Our method also automatically finds a suitable seed by novel centrality measure which identifies ``similarity hubs'' in the graph. The proposed approach in the unsupervised manner outperforms the state-of-the-art methods with classes from the popular benchmark datasets.

In this paper, we propose a novel unsupervised feature selection method: Simultaneous Orthogonal basis Clustering Feature Selection (SOCFS). To perform feature selection on unlabeled data effectively, a regularized regression-based formulation with a new type of target matrix is designed. The target matrix captures latent cluster centers of the projected data points by performing orthogonal basis clustering, and then guides the projection matrix to select discriminative features. Unlike the recent unsupervised feature selection methods, SOCFS does not explicitly use the pre-computed local structure information for data points represented as additional terms of their objective functions, but directly computes latent cluster information by the target matrix conducting orthogonal basis clustering in a single unified term of the proposed objective function. It turns out that the proposed objective function can be minimized by a simple optimization algorithm. Experimental results demonstrate the effectiveness of SOCFS achieving the state-of-the-art results with diverse real world datasets.

This paper addresses unsupervised discovery and localization of dominant objects from a noisy image collection with multiple object classes. The setting of this problem is fully unsupervised, without even image-level annotations or any assumption of a single dominant class. This is far more general than typical colocalization, cosegmentation, or weakly-supervised localization tasks. We tackle the discovery and localization problem using a part-based region matching approach: We use off-the-shelf region proposals to form a set of candidate bounding boxes for objects and object parts. These regions are efficiently matched across images using a probabilistic Hough transform that evaluates the confidence for each candidate correspondence considering both appearance and spatial consistency. Dominant objects are discovered and localized by comparing the scores of candidate regions and selecting those that stand out over other regions containing them. Extensive experimental evaluations on standard benchmarks demonstrate that the proposed approach significantly outperforms the current state of the art in colocalization, and achieves robust object discovery in challenging mixed-class datasets.

In this paper we present a novel framework for unsupervised kinematic structure learning of complex articulated objects from a single-view image sequence. In contrast to prior motion information based methods, which estimate relatively simple articulations, our method can generate arbitrarily complex kinematic structures with skeletal topology by a successive iterative merge process. The iterative merge process is guided by a skeleton distance function which is generated from a novel object boundary generation method from sparse points. Our main contributions can be summarised as follows: (i) Unsupervised complex articulated kinematic structure learning by combining motion and skeleton information. (ii) Iterative fine-to-coarse merging strategy for adaptive motion segmentation and structure smoothing. (iii) Skeleton estimation from sparse feature points. (iv) A new highly articulated object dataset containing multi-stage complexity with ground truth. Our experiments show that the proposed method out-performs state-of-the-art methods both quantitatively and qualitatively.

The construction and fitting of deformable models that capture the degrees of freedom of articulated objects is one of the most popular areas of research in computer vision. The two main approaches are: Holistic Deformable Models (HDMs), which try to represent the object as a whole, and Parts-Based Deformable Models (PBDMs), which model object parts independently. Both models have their own advantages. In this paper we try to marry the previous two frameworks into a unified one that potentially combines the advantages of both. We do so by merging the popular Active Appearance Models (holistic) and Constrained Local Models (part-based) using a novel probabilistic formulation of the fitting problem. To the best of our knowledge, this is the first time that such an idea has been proposed. We show that our unified holistic and part-based formulation achieves state-of-the-art results in the problem of face alignment in-the-wild. Finally, in order to encourage open research and facilitate future comparisons with the proposed method, our code will be made publicly available to the research community.

Marginal histograms provide valuable information for various computer vision problems. However, current image restoration methods do not fully exploit the potential of marginal histograms, in particular, their role as ensemble constraints on the marginal statistics of the restored image. In this paper, we introduce a new framework, UniHIST, to incorporate marginal histogram constraints into image restoration. The key idea of UniHIST is to minimize the discrepancy between the marginal histograms of the restored image and the reference histograms in pixel or gradient domains using the quadratic Wasserstein (W2) distance. The W2 distance can be computed directly from data without resorting to density estimation. It provides a differentiable metric between marginal histograms and allows easy integration with existing image restoration methods. We demonstrate the effectiveness of UniHIST through denoising of pattern images and non-blind deconvolution of natural images. We show that UniHIST enhances restoration performance and leads to visual and quantitative improvements over existing state-of-the-art methods.

In this paper, we present a new framework - task-oriented modeling, learning and recognition which aims at understanding the underlying functions, physics and causality in using objects as "tools". Given a task, such as, cracking a nut or painting a wall, we represent each object, e.g. a hammer or brush, in a generative spatio-temporal representation consisting of four components: i) an affordance basis to be grasped by hand; ii) a functional basis to act on a target object (the nut), iii) the imagined actions with typical motion trajectories; and iv) the underlying physical concepts, e.g. force, pressure, etc. In a learning phase, our algorithm observes only one RGB-D video, in which a rational human picks up one object (i.e. tool) among a number of candidates to accomplish the task. From this example, our algorithm learns the essential physical concepts in the task (e.g. forces in cracking nuts). In an inference phase, our algorithm is given a new set of objects (daily objects or stones), and picks the best choice available together with the inferred affordance basis, functional basis, imagined human actions (sequence of poses), and the expected physical quantity that it will produce. From this new perspective, any objects can be viewed as a hammer or a shovel, and object recognition is not merely memorizing typical appearance examples for each category but reasoning the physical mechanisms in various tasks to achieve generalization.

Pedestrian behavior modeling and analysis is important for crowd scene understanding and has various applications in video surveillance. Stationary crowd groups are a key factor influencing pedestrian walking patterns but was largely ignored in literature. In this paper, a novel model is proposed for pedestrian behavior modeling by including stationary crowd groups as a key component. Through inference on the interactions between stationary crowd groups and pedestrians, our model can be used to investigate pedestrian behaviors. The effectiveness of the proposed model is demonstrated through multiple applications, including walking path prediction, destination prediction, personality classification, and abnormal event detection. To evaluate our model, a large pedestrian walking route dataset is built. The walking routes of 12, 684 pedestrians from a one-hour crowd surveillance video are manually annotated. It will be released to the public and benefit future research on pedestrian behavior analysis and crowd scene understanding.

Virality of online content on social networking websites is an important but esoteric phenomenon often studied in fields like marketing, psychology and data mining. In this paper we study viral images from a computer vision perspective. We introduce three new image datasets from Reddit, and define a virality score using Reddit metadata. We train classifiers with state-of-the-art image features to predict virality of individual images, relative virality in pairs of images, and the dominant topic of a viral image. We also compare machine performance to human performance on these tasks. We find that computers perform poorly with low level features, and high level information is critical for predicting virality. We encode semantic information through relative attributes. We identify the 5 key visual attributes that correlate with virality. We create an attribute-based characterization of images that can predict relative virality with 68.10% accuracy (SVM+Deep Relative Attributes) -- better than humans at 60.12%. Finally, we study how human prediction of image virality varies with different `contexts' in which the images are viewed, such as the influence of neighbouring images, images recently viewed, as well as the image title or caption. This work is a first step in understanding the complex but important phenomenon of image virality. Our datasets and annotations will be made publicly available.

Exploring image structure is a long-standing yet important research subject in the computer vision community. In this paper, we focus on understanding image structure inspired by the "simple-to-complex" biological evidence. A hierarchical shape parsing strategy is proposed to partition and organize image components into a hierarchical structure in the scale space. To improve the robustness and flexibility of image representation, we further bundle the image appearances into hierarchical parsing trees. Image descriptions are subsequently constructed by performing a structural pooling, facilitating efficient matching between the parsing trees. We leverage the proposed hierarchical shape parsing to study two exemplar applications including edge scale refinement and unsupervised "objectness" detection. We show competitive parsing performance comparing to the state-of-the-arts in above scenarios with far less proposals, which thus demonstrates the advantage of the proposed parsing scheme.

Despite the importance of image representations such as histograms of oriented gradients and deep Convolutional Neural Networks (CNN), our theoretical understanding of them remains limited. Aiming at filling this gap, we investigate three key mathematical properties of representations: equivariance, invariance, and equivalence. Equivariance studies how transformations of the input image are encoded by the representation, invariance being a special case where a transformation has no effect. Equivalence studies whether two representations, for example two different parametrisations of a CNN, capture the same visual information or not. A number of methods to establish these properties empirically are proposed, including introducing transformation and stitching layers in CNNs. These methods are then applied to popular representations to reveal insightful aspects of their structure, including clarifying at which layers in a CNN certain geometric invariances are achieved. While the focus of the paper is theoretical, direct applications to structured-output regression are demonstrated too.

Image representations, from SIFT and Bag of Visual Words to Convolutional Neural Networks (CNNs), are a crucial component of almost any image understanding system. Nevertheless, our understanding of them remains limited. In this paper we conduct a direct analysis of the visual information contained in representations by asking the following question: given an encoding of an image, to which extent is it possible to reconstruct the image itself? To answer this question we contribute a general framework to invert representations. We show that this method can invert representations such as HOG more accurately than recent alternatives while being applicable to CNNs too. We then use this technique to study the inverse of recent state-of-the-art CNN image representations for the first time. Among our findings, we show that several layers in CNNs retain photographically accurate information about the image, with different degrees of geometric and photometric invariance.

Modern supervised learning algorithms can learn very accurate and complex discriminating functions. But when these classifiers fail, this complexity can also be a drawback because there is no easy, intuitive way to diagnose why they are failing and remedy the problem. This important question has received little attention. To address this problem, we propose a novel method to analyze and understand a classifier's errors. Our method centers around a measure of how much influence a training example has on the classifier's prediction for a test example. To understand why a classifier is mispredicting the label of a given test example, the user can find and review the most influential training examples that caused this misprediction, allowing them to focus their attention on relevant areas of the data space. This will aid the user in determining if and how the training data is inconsistently labeled or lacking in diversity, or if the feature representation is insufficient. As computing the influence of each training example is computationally impractical, we propose a novel distance metric to approximate influence for boosting classifiers that is fast enough to be used interactively. We also show several novel use paradigms of our distance metric. Through experiments, we show that it can be used to find incorrectly or inconsistently labeled training examples, to find specific areas of the data space that need more training data, and to gain insight into which features are missing from the current representation.

We introduce a realtime facial tracking system specifically designed for performance capture in unconstrained settings using a consumer-level RGB-D sensor. Our framework provides uninterrupted 3D facial tracking, even in the presence of extreme occlusions such as those caused by hair, hand-to-face gestures, and wearable accessories. Anyone's face can be instantly tracked and the users can be switched without an extra calibration step. During tracking, we explicitly segment face regions from any occluding parts by detecting outliers in the shape and appearance input using an exponentially smoothed and user-adaptive tracking model as prior. Our face segmentation combines depth and RGB input data and is also robust against illumination changes. To enable continuous and reliable facial feature tracking in the color channels, we synthesize plausible face textures in the occluded regions. Our tracking model is personalized on-the-fly by progressively refining the user's identity, expressions, and texture with reliable samples and temporal filtering. We demonstrate robust and high-fidelity facial tracking on a wide range of subjects with highly incomplete and largely occluded data. Our system works in everyday environments and is fully unobtrusive to the user, impacting consumer AR applications and surveillance.

This paper presents an algorithm for unconstrained 3D face reconstruction. The input to our algorithm is an "unconstrained" collection of face images captured under a diverse variation of poses, expressions, and illuminations, without meta data about cameras or timing. The output of our algorithm is a true 3D face surface model represented as a watertight triangulated surface with albedo data or texture information. 3D face reconstruction from a collection of unconstrained 2D images is a long-standing computer vision problem. Motivated by the success of the state-of-the-art method, we developed a novel photometric stereo-based method with two distinct novelties. First, working with a true 3D model allows us to enjoy the benefits of using images from all possible poses, including profiles. Second, by leveraging emerging face alignment techniques and our novel normal field-based Laplace editing, a combination of landmark constraints and photometric stereo-based normals drives our surface reconstruction. Given large photo collections and a ground truth 3D surface, we demonstrate the effectiveness and strength of our algorithm both qualitatively and quantitatively.

This paper addresses the problem of uncalibrated photometric stereo with isotropic reflectances. Existing methods face difficulty in solving for the elevation angles of surface normals when the light sources only cover the visible hemisphere. Here, we introduce the notion of "constrained half-vector symmetry" for general isotropic BRDFs and show its capability of elevation angle recovery. This sort of symmetry can be observed in a 1D BRDF slice from a subset of surface normals with the same azimuth angle, and we use it to devise an efficient modeling and solution method to constrain and recover the elevation angles of surface normals accurately. To enable our method to work in an uncalibrated manner, we further solve for light sources in the case of general isotropic BRDFs. By combining this method with the existing ones for azimuth angle estimation, we can get state-of-the-art results for uncalibrated photometric stereo with general isotropic reflectances.

We propose a new vector encoding scheme (tree quantization) that obtains lossy compact codes for high-dimensional vectors via tree-based dynamic programming. Similarly to several previous schemes such as product quantization, these codes correspond to codeword numbers within multiple codebooks. We propose an integer programming-based optimization that jointly recovers the coding tree structure and the codebooks by minimizing the compression error on a training dataset. In the experiments with diverse visual descriptors (SIFT, neural codes, Fisher vectors), tree quantization is shown to combine fast encoding and state-of-the-art accuracy in terms of the compression error, the retrieval performance, and the image classification error.