CVPR 2015 Papers

Many Computer Vision problems arise from information processing of data sources with nuisance variances like scale, orientation, contrast, perspective foreshortening or - in medical imaging - staining and local warping. In most cases these variances can be stated a priori and can be used to improve the generalization of recognition algorithms. We propose a novel supervised feature learning approach, which efficiently extracts information from these constraints to produce interpretable, transformation-invariant features. The proposed method can incorporate a large class of transformations, e.g., shifts, rotations, change of scale, morphological operations, non-linear distortions, photometric transformations, etc. These features boost the discrimination power of a novel image classification and segmentation method, which we call Transformation-Invariant Convolutional Jungles (TICJ). We test the algorithm on two benchmarks in face recognition and medical imaging, where it achieves state of the art results, while being computationally significantly more efficient than Deep Neural Networks.

This paper presents a generic method for transforming MRFs for the marginal inference problem. Its major application is to downsize MRFs to speed up the computation. Unlike the MAP inference, there are only classical algorithms for the marginal inference problem such as BP etc. that require large computational cost. Although downsizing MRFs should directly reduce the computational cost, there is no systematic way of doing this, since it is unclear how to obtain the MRF energy for the downsized MRFs and also how to translate the estimates of their marginal distributions to those of the original MRFs. The proposed method resolves these issues by a novel probabilistic formulation of MRF transformation. The key idea is to represent the joint distribution of an MRF with that of the transformed one, in which the variables of the latter are treated as latent variables. We also show that the proposed method can be applied to discretization of variable space of continuous MRFs and can be used with Markov chain Monte Carlo methods. The experimental results demonstrate the effectiveness of the proposed method.

Learning semantic attributes for person re-identification and description-based person search has gained increasing interest due to attributes' great potential as a pose and view-invariant representation. However, existing attribute-centric approaches have thus far underperformed state-of-the-art conventional approaches. This is due to their non-scalable need for extensive domain (camera) specific annotation. In this paper we present a new semantic attribute learning approach for person re-identification and search. Our model is trained on existing fashion photography datasets -- either weakly or strongly labelled. It can then be transferred and adapted to provide a powerful semantic description of surveillance person detections, without requiring any surveillance domain supervision. The resulting representation is useful for both unsupervised and supervised person re-identification, achieving state-of-the-art and near state-of-the-art performance respectively. Furthermore, as a semantic representation it allows description-based person search to be integrated within the same framework.

In this paper, we show that the seminal, biologically-inspired saliency model by Itti et al. is still competitive with current state-of-the-art methods for salient object segmentation if some important adaptions are made. We show which changes are necessary to achieve high performance, with special emphasis on the scale-space: we introduce a twin pyramid for computing Difference-of-Gaussians, which enables a flexible center-surround ratio. The resulting system, called VOCUS2, is elegant and coherent in structure, fast, and computes saliency at the pixel level. It is not only suitable for images with few objects, but also for complex scenes as captured by mobile devices. Furthermore, we integrate the saliency system into an object proposal generation framework to obtain segment-based saliency maps and boost the results for salient object segmentation. We show that our system achieves state-of-the-art performance on a large collection of benchmark data.

Depth estimation and semantic segmentation are two fundamental problems in image understanding. While the two tasks are strongly correlated and mutually beneficial, they are usually solved separately or sequentially. Motivated by the complementary properties of the two tasks, we propose a unified framework for joint depth and semantic prediction. Given an image, we first use a trained Convolutional Neural Network (CNN) to jointly predict a global layout composed of pixel-wise depth values and semantic labels. By allowing for interactions between the depth and semantic information, the joint network provides more accurate depth prediction than a state-of-the-art CNN trained solely for depth prediction [5]. To further obtain fine-level details, the image is decomposed into local segments for region-level depth and semantic prediction under the guidance of global layout. Utilizing the pixel-wise global prediction and region-wise local prediction, we formulate the inference problem in a two-layer Hierarchical Conditional Random Field (HCRF) to produce the final depth and semantic map. As demonstrated in the experiments, our approach effectively leverages the advantages of both tasks and provides the state-of-the-art results.

With the of advent rich classification models and high computational power visual recognition systems have found many operational applications. Recognition in the real world poses multiple challenges that are not apparent in controlled lab environments. The datasets are dynamic and novel categories must be continuously detected and then added. At prediction time, a trained system has to deal with myriad unseen categories. Operational systems require minimum down time, even to learn. To handle these operational issues, we present the problem of Open World recognition and formally define it. We prove that thresholding sums of monotonically decreasing functions of distances in linearly transformed feature space can balance open space risk and empirical risk. Our theory extends existing algorithms for open world recognition. We present a protocol for evaluation of open world recognition systems. We present the Nearest Non-Outlier (NNO) algorithm which evolves model efficiently, adding object categories incrementally while detecting outliers and managing open space risk. We perform experiments on the ImageNet dataset with 1.2M+ images to validate the effectiveness of our method on large scale visual recognition tasks. NNO consistently yields superior results on open world recognition.

We present a novel, non-intrusive approach for estimating contact forces during hand-object interactions relying solely on visual input provided by a single RGB-D camera. We consider a manipulated object with known geometrical and physical properties. First, we rely on model-based visual tracking to estimate the object's pose together with that of the hand manipulating it throughout the motion. Following this, we compute the object's first and second order kinematics using a new class of numerical differentiation operators. The estimated kinematics is then instantly fed into a second-order cone program that returns a minimal force distribution explaining the observed motion. However, humans typically apply more forces than mechanically required when manipulating objects. Thus, we complete our estimation method by learning these excessive forces and their distribution among the fingers in contact. We provide a full validity analysis of the proposed method by evaluating it based on ground truth data from additional sensors such as accelerometers, gyroscopes and pressure sensors. Experimental results show that force sensing from vision (FSV) is indeed feasible.

Nowadays, detecting objects in 3D scenes like point clouds has become an emerging challenge with various applications. However, it retains as an open problem due to the deficiency of labeling 3D training data. To deploy an accurate detection algorithm typically resorts to investigating both RGB and depth modalities, which have distinct statistics while correlated with each other. Previous research mainly focus on detecting objects using only one modality, which ignores exploiting the cross-modality cues. In this work, we propose a cross-modality deep learning framework based on deep Boltzmann Machines for 3D Scenes object detection. In particular, we demonstrate that by learning cross-modality feature from RGBD data, it is possible to capture their joint information to reinforce detector trainings in individual modalities. In particular, we slide a 3D detection window in the 3D point cloud to match the exemplar shape, which the lack of training data in 3D domain is conquered via (1) We collect 3D CAD models and 2D positive samples from Internet. (2) adopt pretrained R-CNNs [2] to extract raw feature from both RGB and Depth domains. Experiments on RMRC dataset demonstrate that the bimodal based deep feature learning framework helps 3D scene object detection.

Human shape tracking consists in fitting a template model to temporal sequences of visual observations. It usually comprises an association step, that finds correspondences between the model and the input data, and a deformation step, that fits the model to the observations given correspondences. Most current approaches find their common ground with the Iterative-Closest-Point (ICP) algorithm, which facilitates the association step with local distance considerations. It fails when large deformations occur, and errors in the association tend to propagate over time. In this paper, we propose a discriminative alternative for the association, that leverages random forests to infer correspondences in one shot. It allows for large deformations and prevents tracking errors from accumulating. The approach is successfully integrated to a surface tracking framework that recovers human shapes and poses jointly. When combined with ICP, this discriminative association proves to yield better accuracy in registration, more stability when tracking over time, and faster convergence. Evaluations on existing datasets demonstrate the benefits with respect to the state-of-the-art.

This paper introduces the concept of shape signals, i.e., series of shapes which have a natural temporal or spatial ordering, as well as a variational formulation for the regularization of these signals. The proposed formulation can be seen as the shape-valued generalization of the Rudin-Osher-Fatemi (ROF) functional for intensity images. We derive a variant of the classical finite-dimensional representation of Kendall, but our framework is generic in the sense that it can be combined with any shape space. This representation allows for the explicit computation of geodesics and thus facilitates the efficient numerical treatment of the variational formulation by means of the cyclic proximal point algorithm. Similar to the ROF-functional, we demonstrate experimentally that l_1-type penalties both for data fidelity term and regularizer perform best in regularizing shape signals. Finally, we show applications of our method to shape signals obtained from synthetic, photometric, and medical data sets.

It is known that time-to-contact toward objects can be estimated just from changes in the object size in camera images, and we do not need any additional information, such as camera parameters and motions. However, the existing methods for measuring the time-to-contact are based on geometric image features, such as corners and edge lines, and thus they cannot be used when there are no geometric features in images. In this paper, we propose a new method for computing the time-to-contact from photometric information in images. When a light source moves in the 3D scene, an observed intensity changes according to the motion of the light source. In this paper, we analyze the change in photometric information in images, and show that the time-to-contact can be estimated just from the changes in intensity in images. Our method does not need any additional information, such as radiance of light source, reflectance of object and orientation of object surface. The proposed method can be used in various applications, such as vehicle driver assistance.

In contrast to category-level or cluster-level classifiers, exemplar SVM is successfully applied to classifying (or detecting) a target object as well as transferring instance-level annotations. The method, however, is formulated in a highly biased classification problem where only one positive sample is contrasted with a substantial number of negative samples, which makes it difficult to properly determine the regularization parameters balancing two types of costs derived from positive and negative samples. In this paper, we present two novel viewpoints toward exemplar SVM in addition to the original definition. From these proposed viewpoints, we can give light on an intrinsic structure of exemplar SVM, reducing two parameters into only one as well as providing clear intuition on the parameter, in order to free us from exhaustive parameter tuning. We can also clarify how the classifier geometrically works so as to produce homogeneous classification scores of multiple exemplar SVMs which are comparable to each other without calibration. In addition, we propose a novel feature transformation method based on those viewpoints which contributes to general classification tasks. In the experiments on object detection and image classification, the proposed methods regarding exemplar SVM exhibit favorable performance.

Sparsity, or cardinality, as a tool for feature selection is extremely common in a vast number of current computer vision applications. The $k$-support norm is a recently proposed norm with the proven property of providing the tightest convex bound on cardinality over the Euclidean norm unit ball. In this paper we present a re-derivation of this norm, with the hope of shedding further light on this particular surrogate function. In addition, we also present a connection between the rank operator, the nuclear norm and the $k$-support norm. Finally, based on the results established in this re-derivation, we propose a novel algorithm with significantly improved computational efficiency, empirically validated on a number of different problems, using both synthetic and real world data.

A number of recent studies have shown that a Deep Convolutional Neural Network (DCNN) pretrained on a large dataset can be adopted as a universal image descriptor, and that doing so leads to impressive performance at a range of image classification tasks. Most of these studies, if not all, adopt activations of the fully-connected layer of a DCNN as the image or region representation and it is believed that convolutional layer activations are less discriminative. This paper, however, advocates that if used appropriately, convolutional layer activations constitute a powerful image representation. This is achieved by adopting a new technique proposed in this paper called cross-convolutional-layer pooling. More specifically, it extracts subarrays of feature maps of one convolutional layer as local features, and pools the extracted features with the guidance of the feature maps of the successive convolutional layer. Compared with existing methods that apply DCNNs in the similar local feature setting, the proposed method avoids the input image style mismatching issue which is usually encountered when applying fully connected layer activations to describe local regions. Also, the proposed method is easier to implement since it is codebook free and does not have any tuning parameters. By applying our method to four popular visual classification tasks, it is demonstrated that the proposed method can achieve comparable or in some cases significantly better performance than existing fully-connected layer based image representations.

We propose a new 3D model of the human body that is both realistic and part-based. The body is represented by a graphical model in which nodes of the graph correspond to body parts that can independently translate and rotate in 3D and deform to represent different body shapes and to capture pose-dependent shape variations. Pairwise potentials define a "stitching cost" for pulling the limbs apart, giving rise to the Stitched Puppet (SP) model. Unlike existing realistic 3D body models, the distributed representation facilitates inference by allowing the model to more effectively explore the space of poses, much like existing 2D pictorial structures models. We infer pose and body shape using a form of particle-based max-product belief propagation. This gives SP the realism of recent 3D body models with the computational advantages of part-based models. We apply SP to two challenging problems involving estimating human shape and pose from 3D data. The first is the FAUST mesh alignment challenge, where ours is the first method to successfully align all 3D meshes with no pose prior. The second involves estimating pose and shape from crude visual hull representations of complex body movements.

The topic of crowd modeling in computer vision usually assumes a single generic typology of crowd, which is very simplistic. In this paper we adopt a taxonomy that is widely accepted in sociology, focusing on a particular category, the spectator crowd, which is formed by people "interested in watching something specific that they came to see". This can be found at the stadiums, amphitheaters, cinema, etc. In particular, we propose a novel dataset, the Spectators Hockey (S-Hock), which deals with 4 hockey matches during an international tournament. In the dataset, a massive annotation has been carried out, focusing on the spectators at different levels of details: at a higher level, people have been labeled depending on the team they are supporting and the fact that they know the people close to them; going to the lower levels, standard pose information has been considered (regarding the head, the body) but also fine grained actions such as hands on hips, clapping hands etc. The labeling focused on the game field also, permitting to relate what is going on in the match with the crowd behavior. This brought to more than 100 millions of annotations, useful for standard applications as people counting and head pose estimation but also for novel tasks as spectator categorization. For all of these we provide protocols and baseline results, encouraging further research.

In projective geometry, the common self-polar triangle has often been used to discuss the position relationship of two planar conics. However, there are few researches on the properties of the common self-polar triangle, especially when the two planar conics are special conics. In this paper, we explore the properties of the common self-polar triangle, when the two conics happen to be concentric circles. We show there exist infinite many common self-polar triangles of two concentric circles, and provide a method to locate the vertices of these triangles. By investigating all these triangles, we find that they encode two important properties. The first one is all triangles share one common vertex, and the opposite side of the common vertex lies on the same line, which are the circle center and the line at the infinity of the support plane. The second is all triangles are right triangles. Based on these two properties, the imaged circle center and the varnishing line of support plane can be recovered simultaneously, and many conjugate pairs on vanishing line can be obtained. These allow to induce good constraints on the image of absolute conic. We evaluate two calibration algorithms, whereby accurate results are achieved. The main contribution of this paper is that we initiate a new perspective to look into circle-based camera calibration problem. We believe that other calibration methods using different circle patterns can benefit from this perspective, especially for the patterns which involve more than two circles.

Fine-grained classification is challenging because categories can only be discriminated by subtle and local differences. Variances in the pose, scale or rotation usually make the problem more difficult. Most fine-grained classification systems follow the pipeline of finding foreground object or object parts (where) to extract discriminative features (what). In this paper, we propose to apply visual attention to fine-grained classification task using deep neural network. Our pipeline integrates three types of attention: the bottom-up attention that propose candidate patches, the object-level top-down attention that selects relevant patches to a certain object, and the part-level top-down attention that localizes discriminative parts. We combine these attentions to train domain-specific deep nets, then use it to improve both the what and where aspects. Importantly, we avoid using expensive annotations like bounding box or part information from end-to-end. The weak supervision constraint makes our work easier to generalize. We have verified the effectiveness of the method on the subsets of ILSVRC2012 dataset and CUB200_2011 dataset. Our pipeline delivered significant improvements and achieved the best accuracy under the weakest supervision condition. The performance is competitive against other methods that rely on additional annotations.

When viewed through a small aperture, a moving image provides incomplete information about the local motion. Only the component of motion along the local image gradient is constrained. In an essential part of optical flow algorithms, information must be aggregated from nearby image locations in order to estimate all components of motion. This limitation of local evidence for estimating optical flow is called ''the aperture problem''. We pose and solve a generalization of the aperture problem for moving refractive elements. We consider a common setup in air flow imaging or telescope observation: a camera is viewing a static background, and an unknown refractive elements undergoing unknown motion between them. Then we are addressing this fundamental question: what does the local image motion tell us about the motion of refractive elements? We show that the information gleaned through a local aperture for this case is very different than that for optical flow. In optical flow, the movement of 1D structure already constrains the motion in a certain direction. However, we cannot infer any information about the refractive motion from the movement of 1D structure in the observed sequence, and can only recover one component of the motion from 2D structure. Results on both simulated and real sequences are shown to illustrate our theory.