ICCV 2015 论文列表

Sparse coding (SC) plays an important role in versatile computer vision applications such as image super-resolution (SR). Most of the previous SC based SR methods partition the image into overlapped patches, and process each patch separately. These methods, however, ignore the consistency of pixels in overlapped patches, which is a strong constraint for image reconstruction. In this paper, we propose a convolutional sparse coding (CSC) based SR (CSC-SR) method to address the consistency issue. Our CSC-SR involves three groups of parameters to be learned: (i) a set of filters to decompose the low resolution (LR) image into LR sparse feature maps; (ii) a mapping function to predict the high resolution (HR) feature maps from the LR ones; and (iii) a set of filters to reconstruct the HR images from the predicted HR feature maps via simple convolution operations. By working directly on the whole image, the proposed CSC-SR algorithm does not need to divide the image into overlapped patches, and can exploit the image global correlation to produce more robust reconstruction of image local structures. Experimental results clearly validate the advantages of CSC over patch based SC in SR application. Compared with state-of-the-art SR methods, the proposed CSC-SR method achieves highly competitive PSNR results, while demonstrating better edge and texture preservation performance.

Color constancy is the problem of inferring the color of the light that illuminated a scene, usually so that the illumination color can be removed. Because this problem is underconstrained, it is often solved by modeling the statistical regularities of the colors of natural objects and illumination. In contrast, in this paper we reformulate the problem of color constancy as a 2D spatial localization task in a log-chrominance space, thereby allowing us to apply techniques from object detection and structured prediction to the color constancy problem. By directly learning how to discriminate between correctly white-balanced images and poorly white-balanced images, our model is able to improve performance on standard benchmarks by nearly 40%.

Deep learning methods are powerful tools but often suffer from expensive computation and limited flexibility. An alternative is to combine light-weight models with deep representations. As successful cases exist in several visual problems, a unified framework is absent. In this paper, we revisit two widely used approaches in computer vision, namely filtered channel features and Convolutional Neural Networks (CNN), and absorb merits from both by proposing an integrated method called Convolutional Channel Features (CCF). CCF transfers low-level features from pre-trained CNN models to feed the boosting forest model. With the combination of CNN features and boosting forest, CCF benefits from the richer capacity in feature representation compared with channel features, as well as lower cost in computation and storage compared with end-to-end CNN methods. We show that CCF serves as a good way of tailoring pre-trained CNN models to diverse tasks without fine-tuning the whole network to each task by achieving state-of-the-art performances in pedestrian detection, face detection, edge detection and object proposal generation.

We introduce a convex optimization modeling framework that transforms a convex optimization problem expressed in a form natural and convenient for the user into an equivalent cone program in a way that preserves fast linear transforms in the original problem. By representing linear functions in the transformation process not as matrices, but as graphs that encode composition of abstract linear operators, we arrive at a matrix-free cone program, i.e., one whose data matrix is represented by an abstract linear operator and its adjoint. This cone program can then be solved by a matrix-free cone solver. By combining the matrix-free modeling framework and cone solver, we obtain a general method for efficiently solving convex optimization problems involving fast linear transforms.

To find the optimal nonlinear separating boundary with maximum margin in the input data space, this paper proposes Contractive Rectifier Networks (CRNs), wherein the hidden-layer transformations are restricted to be contraction mappings. The contractive constraints ensure that the achieved separating margin in the input space is larger than or equal to the separating margin in the output layer. The training of the proposed CRNs is formulated as a linear support vector machine (SVM) in the output layer, combined with two or more contractive hidden layers. Effective algorithms have been proposed to address the optimization challenges arising from contraction constraints. Experimental results on MNIST, CIFAR-10, CIFAR-100 and MIT-67 datasets demonstrate that the proposed contractive rectifier networks consistently outperform their conventional unconstrained rectifier network counterparts.

For its simplicity and effectiveness, star model is popular in shape matching. However, it suffers from the loose geometric connections among parts. In the paper, we present a novel algorithm that reconsiders these connections and reduces the global matching to a set of interrelated local matching. For the purpose, we divide the shape template into overlapped parts and model the matching through a part-based layered structure that uses the latent variable to constrain parts' deformation. As for inference, each part is used for localizing candidates by the partial matching. Thanks to the contour fragments, the partial matching can be solved via modified dynamic programming. The overlapped regions among parts of the template are then explored to make the candidates of parts meet at their shared points. The process is fulfilled via a refined procedure based on iterative dynamic programming. Results on ETHZ shape and Inria Horse datasets demonstrate the benefits of the proposed algorithm.

Our goal is to estimate contour flow (the contour pairs with consistent point correspondence) from inconsistent contours extracted independently in two video frames. We formulate the contour flow estimation locally as a motion segmentation problem where motion patterns grouped from optical flow field are exploited for local correspondence measurement. To solve local ambiguities, contour flow estimation is further formulated globally as a contour alignment problem. We propose a novel two-staged strategy to obtain global consistent point correspondence under various contour transitions such as splitting, merging and branching. The goal of the first stage is to obtain possible accurate contour-to-contour alignments, and the second stage aims to make a consistent fusion of many partial alignments. Such a strategy can properly balance the accuracy and the consistency, which enables a middle-level motion representation to be constructed by just concatenating frame-by-frame contour flow estimation. Experiments prove the effectiveness of our method.

The bio-inspired, asynchronous event-based dynamic vision sensor records temporal changes in the luminance of the scene at high temporal resolution. Since events are only triggered at significant luminance changes, most events occur at the boundary of objects and their parts. The detection of these contours is an essential step for further interpretation of the scene. This paper presents an approach to learn the location of contours and their border ownership using Structured Random Forests on event-based features that encode motion, timing, texture, and spatial orientations. The classifier integrates elegantly information over time by utilizing the classification results previously computed. Finally, the contour detection and boundary assignment are demonstrated in a layer-segmentation of the scene. Experimental results demonstrate good performance in boundary detection and segmentation.

Closed contour is an important objectness indicator. We propose a new measure subject to the completeness and tightness constraints, where the optimized closed contour should be tightly bounded within an object proposal. The closed contour measure is defined using closed path integral, and we solve the optimization problem efficiently in polar coordinate system with a global optimum guaranteed. Extensive experiments show that our method can reject a large number of false proposals, and achieve over 6% improvement in object recall at the challenging overlap threshold 0.8 on the VOC 2007 test dataset.

In this paper, we treat the problem of continuous pose estimation for object categories as a regression problem on the basis of only 2D training information. While regression is a natural framework for continuous problems, regression methods so far achieved inferior results with respect to 3D-based and 2D-based classification-and-refinement approaches. This may be attributed to their weakness to high intra-class variability as well as to noisy matching procedures and lack of geometrical constraints. We propose to apply regression to Fisher-encoded vectors computed from large cells by learning an array of Fisher regressors. Fisher encoding makes our algorithm flexible to variations in class appearance, while the array structure permits to indirectly introduce spatial context information in the approach. We formulate our problem as a MAP inference problem, where the likelihood function is composed of a generative term based on the prediction error generated by the ensemble of Fisher regressors as well as a discriminative term based on SVM classifiers. We test our algorithm on three publicly available datasets that envisage several difficulties, such as high intra-class variability, truncations, occlusions, and motion blur, obtaining state-of-the-art results.

There are multiple cues in an image which reveal what action a person is performing. For example, a jogger has a pose that is characteristic for jogging, but the scene (e.g. road, trail) and the presence of other joggers can be an additional source of information. In this work, we exploit the simple observation that actions are accompanied by contextual cues to build a strong action recognition system. We adapt RCNN to use more than one region for classification while still maintaining the ability to localize the action. We call our system R*CNN. The action-specific models and the feature maps are trained jointly, allowing for action specific representations to emerge. R*CNN achieves 90.2% mean AP on the PASAL VOC Action dataset, outperforming all other approaches in the field by a significant margin. Last, we show that R*CNN is not limited to action recognition. In particular, R*CNN can also be used to tackle fine-grained tasks such as attribute classification. We validate this claim by reporting state-of-the-art performance on the Berkeley Attributes of People dataset.

Existing approaches for diffusion on graphs, e.g., for label propagation, are mainly focused on isotropic diffusion, which is induced by the commonly-used graph Laplacian regularizer. Inspired by the success of diffusivity tensors for anisotropic diffusion in image processing, we presents anisotropic diffusion on graphs and the corresponding label propagation algorithm. We develop positive definite diffusivity operators on the vector bundles of Riemannian manifolds, and discretize them to diffusivity operators on graphs. This enables us to easily define new robust diffusivity operators which significantly improve semi-supervised learning performance over existing diffusion algorithms.

Person detection is a key problem for many computer vision tasks. While face detection has reached maturity, detecting people under full variation of camera view-points, human poses, lighting conditions and occlusions is still a difficult challenge. In this work we focus on detecting human heads in natural scenes. Starting from the recent R-CNN object detector, we extend it in two ways. First, we leverage person-scene relations and propose a global CNN model trained to predict positions and scales of heads directly from the full image. Second, we explicitly model pairwise relations among the objects via energy-based model where the potentials are computed with a CNN framework. Our full combined model complements R-CNN with contextual cues derived from the scene. To train and test our model, we introduce a large dataset with 369,846 human heads annotated in 224,740 movie frames. We evaluate our method and demonstrate improvements of person head detection compared to several recent baselines on three datasets. We also show improvements of the detection speed provided by our model.

Activity recognition in video has recently benefited from the use of the context e.g., inter-relationships among the activities and objects. However, these approaches require data to be labeled and entirely available at the outset. In contrast, we formulate a continuous learning framework for context aware activity recognition from unlabeled video data which has two distinct advantages over most existing methods. First, we propose a novel active learning technique which not only exploits the informativeness of the individual activity instances but also utilizes their contextual information during the query selection process; this leads to significant reduction in expensive manual annotation effort. Second, the learned models can be adapted online as more data is available. We formulate a conditional random field (CRF) model that encodes the context and devise an information theoretic approach that utilizes entropy and mutual information of the nodes to compute the set of most informative query instances, which need to be labeled by a human. These labels are combined with graphical inference techniques for incrementally updating the model as new videos come in. Experiments on four challenging datasets demonstrate that our framework achieves superior performance with significantly less amount of manual labeling.

We present an approach to learn a dense pixel-wise labeling from image-level tags. Each image-level tag imposes constraints on the output labeling of a Convolutional Neural Network (CNN) classifier. We propose Constrained CNN (CCNN), a method which uses a novel loss function to optimize for any set of linear constraints on the output space (i.e. predicted label distribution) of a CNN. Our loss formulation is easy to optimize and can be incorporated directly into standard stochastic gradient descent optimization. The key idea is to phrase the training objective as a biconvex optimization for linear models, which we then relax to nonlinear deep networks. Extensive experiments demonstrate the generality of our new learning framework. The constrained loss yields state-of-the-art results on weakly supervised semantic image segmentation. We further demonstrate that adding slightly more supervision can greatly improve the performance of the learning algorithm.

Obtaining an appropriate dictionary is the key point when sparse representation is applied to computer vision or image processing problems such as image restoration. It is expected that preserving data structure during sparse coding and dictionary learning can enhance the recovery performance. However, many existing dictionary learning methods handle training samples individually, while missing relationships between samples, which result in dictionaries with redundant atoms but poor representation ability. In this paper, we propose a novel sparse representation approach called conformal and low-rank sparse representation (CLRSR) for image restoration problems. To achieve a more compact and representative dictionary, conformal property is introduced by preserving the angles of local geometry formed by neighboring samples in the feature space. Furthermore, imposing low-rank constraint on the coefficient matrix can lead more faithful subspaces and capture the global structure of data. We apply our CLRSR model to several image restoration tasks to demonstrate the effectiveness.

Varied sources of error contribute to the challenge of facial action unit detection. Previous approaches address specific and known sources. However, many sources are unknown. To address the ubiquity of error, we propose a Confident Preserving Machine (CPM) that follows an easy-to-hard classification strategy. During training, CPM learns two confident classifiers. A confident positive classifier separates easily identified positive samples from all else; a confident negative classifier does same for negative samples. During testing, CPM then learns a person-specific classifier using ``virtual labels'' provided by confident classifiers. This step is achieved using a quasi-semi-supervised (QSS) approach. Hard samples are typically close to the decision boundary, and the QSS approach disambiguates them using spatio-temporal constraints. To evaluate CPM, we compared it with a baseline single-margin classifier and state-of-the-art semi-supervised learning, transfer learning, and boosting methods in three datasets of spontaneous facial behavior. With few exceptions, CPM outperformed baseline and state-of-the art methods.

Single image super-resolution is an important task in the field of computer vision and finds many practical applications. Current state-of-the-art methods typically rely on machine learning algorithms to infer a mapping from low- to high-resolution images. These methods use a single fixed blur kernel during training and, consequently, assume the exact same kernel underlying the image formation process for all test images. However, this setting is not realistic for practical applications, because the blur is typically different for each test image. In this paper, we loosen this restrictive constraint and propose conditioned regression models (including convolutional neural networks and random forests) that can effectively exploit the additional kernel information during both, training and inference. This allows for training a single model, while previous methods need to be re-trained for every blur kernel individually to achieve good results, which we demonstrate in our evaluations. We also empirically show that the proposed conditioned regression models (i) can effectively handle scenarios where the blur kernel is different for each image and (ii) outperform related approaches trained for only a single kernel.

Pixel-level labelling tasks, such as semantic segmentation, play a central role in image understanding. Recent approaches have attempted to harness the capabilities of deep learning techniques for image recognition to tackle pixel-level labelling tasks. One central issue in this methodology is the limited capacity of deep learning techniques to delineate visual objects. To solve this problem, we introduce a new form of convolutional neural network that combines the strengths of Convolutional Neural Networks (CNNs) and Conditional Random Fields (CRFs)-based probabilistic graphical modelling. To this end, we formulate Conditional Random Fields with Gaussian pairwise potentials and mean-field approximate inference as Recurrent Neural Networks. This network, called CRF-RNN, is then plugged in as a part of a CNN to obtain a deep network that has desirable properties of both CNNs and CRFs. Importantly, our system fully integrates CRF modelling with CNNs, making it possible to train the whole deep network end-to-end with the usual back-propagation algorithm, avoiding offline post-processing methods for object delineation. We apply the proposed method to the problem of semantic image segmentation, obtaining top results on the challenging Pascal VOC 2012 segmentation benchmark.