CVPR 2016 Papers

Recently, Deep Convolutional Neural Networks (DCNNs) have been applied to the task of human pose estimation, and have shown its potential of learning better feature representations and capturing contextual relationships. However, it is difficult to incorporate domain prior knowledge such as geometric relationships among body parts into DCNNs. In addition, training DCNN-based body part detectors without consideration of global body joint consistency introduces ambiguities, which increases the complexity of training. In this paper, we propose a novel end-to-end framework for human pose estimation that combines DCNNs with the expressive deformable mixture of parts. We explicitly incorporate domain prior knowledge into the framework, which greatly regularizes the learning process and enables the flexibility of our framework for loopy models or tree-structured models. The effectiveness of jointly learning a DCNN with a deformable mixture of parts model is evaluated through intensive experiments on several widely used benchmarks. The proposed approach significantly improves the performance compared with state-of-the-art approaches, especially on benchmarks with challenging articulations.

In this work we introduce a fully end-to-end approach for action detection in videos that learns to directly predict the temporal bounds of actions. Our intuition is that the process of detecting actions is naturally one of observation and refinement: observing moments in video, and refining hypotheses about when an action is occurring. Based on this insight, we formulate our model as a recurrent neural network-based agent that interacts with a video over time. The agent observes video frames and decides both where to look next and whether to emit a prediction. Since backpropagation is not adequate in this non-differentiable setting, we use REINFORCE to learn the agent's task-specific decision policy. Our model achieves state-of-the-art results on the THUMOS'14 and ActivityNet datasets while observing only a fraction (2% or less) of the video frames.

Research in face perception and emotion theory requires very large annotated databases of images of facial expressions of emotion. Annotations should include Action Units (AUs) and their intensities as well as emotion category. This goal cannot be readily achieved manually. Herein, we present a novel computer vision algorithm to annotate a large database of one million images of facial expressions of emotion in the wild (i.e., face images downloaded from the Internet). First, we show that this newly proposed algorithm can recognize AUs and their intensities reliably across databases. To our knowledge, this is the first published algorithm to achieve highly-accurate results in the recognition of AUs and their intensities across multiple databases. Our algorithm also runs in real-time (>30 images/second), allowing it to work with large numbers of images and video sequences. Second, we use WordNet to download 1,000,000 images of facial expressions with associated emotion keywords from the Internet. These images are then automatically annotated with AUs, AU intensities and emotion categories by our algorithm. The result is a highly useful database that can be readily queried using semantic descriptions for applications in computer vision, affective computing, social and cognitive psychology and neuroscience; e.g., "show me all the images with happy faces" or "all images with AU 1 at intensity c."

Recent algorithms in convolutional neural networks (CNN) considerably advance the fine-grained image classification, which aims to differentiate the subtle differences among subordinate classes. However, previous studies have rarely focused on learning a fined-grained and structured feature representation that is able to locate relevant images at different levels of relevance, e.g., discovering cars from the same make or the same model, both of which require high precision. In this paper, we propose two main contributions to tackle this problem. 1) A multi-task learning framework is designed to effectively learn fine-grained feature representations by jointly optimizing both classification and similarity constraints. 2) To model the multi-level relevance, label structures such as hierarchy or shared attributes are seamlessly embedded into the framework by generalizing the triplet loss. Extensive and thorough experiments have been conducted on three fine-grained datasets, i.e., the Stanford car, the car-333, and the food datasets, which contain either hierarchical labels or shared attributes. Our proposed method has achieved very competitive performance, i.e., among state-of-the-art classification accuracy. More importantly, it significantly outperforms previous fine-grained feature representations for image retrieval at different levels of relevance

We presents a method for future localization: to predict plausible future trajectories of ego-motion in egocentric stereo images. Our paths avoid obstacles, move between objects, even turn around a corner into space behind objects. As a byproduct of the predicted trajectories, we discover the empty space occluded by foreground objects. One key innovation is the creation of an EgoRetinal map, akin to an illustrated tourist map, that `rearranges' pixels taking into accounts depth information, the ground plane, and body motion direction, so that it allows motion planning and perception of objects on one image space. We learn to plan trajectories directly on this EgoRetinal map using first person experience of walking around in a variety of scenes. In a testing phase, given an novel scene, we find multiple hypotheses of future trajectories from the learned experience. We refine them by minimizing a cost function that describes compatibility between the obstacles in the EgoRetinal map and trajectories. We quantitatively evaluate our method to show predictive validity and apply to various real world daily activities including walking, shopping, and social interactions.

While many recent hand pose estimation methods critically rely on a training set of labelled frames, the creation of such a dataset is a challenging task that has been overlooked so far. As a result, existing datasets are limited to a few sequences and individuals, with limited accuracy, and this prevents these methods from delivering their full potential. We propose a semi-automated method for efficiently and accurately labeling each frame of a hand depth video with the corresponding 3D locations of the joints: The user is asked to provide only an estimate of the 2D reprojections of the visible joints in some reference frames, which are automatically selected to minimize the labeling work by efficiently optimizing a sub-modular loss function. We then exploit spatial, temporal, and appearance constraints to retrieve the full 3D poses of the hand over the complete sequence. We show that this data can be used to train a recent state-of-the-art hand pose estimation method, leading to increased accuracy.

We present an efficient technique to optimize color consistency of a collection of images depicting a common scene. Our method first recovers sparse pixel correspondences in the input images and stacks them into a matrix with many missing entries. We show that this matrix satisfies a rank two constraint under a simple color correction model. These parameters can be viewed as pseudo white balance and gamma correction parameters for each input image. We present a robust low-rank matrix factorization method to estimate the unknown parameters of this model. Using them, we improve color consistency of the input images or perform color transfer with any input image as the source. Our approach is insensitive to outliers in the pixel correspondences thereby precluding the need for complex pre-processing steps. We demonstrate high-quality color consistency results on large photo collections of popular tourist landmarks and personal photo collections containing images of people.

In this paper, we propose training very deep neural networks (DNNs) for supervised learning of hash codes. Existing methods in this context train relatively "shallow" networks limited by the issues arising in back propagation (e.g. vanishing gradients) as well as computational efficiency. We propose a novel and efficient training algorithm inspired by alternating direction method of multipliers (ADMM) that overcomes some of these limitations. Our method decomposes the training process into independent layer-wise local updates through auxiliary variables. Empirically we observe that our training algorithm always converges and its computational complexity is linearly proportional to the number of edges in the networks. Empirically we manage to train DNNs with 64 hidden layers and 1024 nodes per layer for supervised hashing in about 3 hours using a single GPU. Our proposed very deep supervised hashing (VDSH) method significantly outperforms the state-of-the-art on several benchmark datasets.

In this paper we propose a new framework to compare and classify temporal sequences. The proposed approach captures the underlying dynamics of the data while avoiding expensive estimation procedures, making it suitable to process large numbers of sequences. The main idea is to first embed the sequences into a Riemannian manifold by using positive definite regularized Gram matrices of their Hankelets. The advantages of the this approach are: 1) it allows for using non-Euclidean similarity functions on the Positive Definite matrix manifold, which capture better the underlying geometry than directly comparing the sequences or their Hankel matrices; and 2) Gram matrices inherit desirable properties from the underlying Hankel matrices: their rank measure the complexity of the underlying dynamics, and the rank and the coefficients of the associated regressive models are invariant to affine transformations and varying initial conditions. The benefits of this approach are illustrated with extensive experiments in 3D action recognition using 3D joints sequences. In spite of its simplicity, the performance of this approach is competitive or better than using state-of-art approaches for this problem. Further, these results hold across a variety of metrics, supporting the idea that the improvement stems from the embedding itself, rather than from using one of these metrics.

We tackle the problem of large-scale object detection in images, where the number of objects can be arbitrarily large, and can exhibit significant overlap/occlusion. A successful approach to modelling the large-scale nature of this problem has been via point process density functions which jointly encode object qualities and spatial interactions. But the corresponding optimisation problem is typically difficult or intractable, and many of the best current methods rely on Monte Carlo Markov Chain (MCMC) simulation, which converges slowly in a large solution space. We propose an efficient point process inference for large-scale object detection using discrete energy minimization. In particular, we approximate the solution space by a finite set of object proposals and cast the point process density function to a corresponding energy function of binary variables whose values indicate which object proposals are accepted. We resort to the local submodular approximation (LSA) based trust-region optimisation to find the optimal solution. Furthermore we analyse the error of LSA approximation, and show how to adjust the point process energy to dramatically speed up the convergence without harms in the optimality. We demonstrate the superior efficiency and accuracy of our method using a variety of large-scale object detection applications such as crowd human detection, birds, cells counting/localization.

Recent advances in semantic image segmentation have mostly been achieved by training deep convolutional neural networks(CNNs). We show how to improve semantic segmentation through the use of contextual information; specifically, we explore 'patch-patch' context between image regions, and 'patch-background' context. For learning from the patch-patch context, we formulate Conditional Random Fields (CRFs) with CNN-based pairwise potential functions to capture semantic correlations between neighboring patches. Efficient piecewise training of the proposed deep structured model is then applied to avoid repeated expensive CRF inference for back propagation. For capturing the patch-background context, we show that a network design with traditional multi-scale image input and sliding pyramid pooling is effective for improving performance. Our experimental results set new state-of-the-art performance on a number of popular semantic segmentation datasets, including NYUDv2, PASCAL VOC 2012, PASCAL-Context, and SIFT-flow. In particular, we achieve an intersection-over-union score of 78.0 on the challenging PASCAL VOC 2012 dataset.

We consider the image retrieval problem of finding the images in a dataset that are most similar to a query image. Our goal is to reduce the number of vector operations and memory for performing a search without sacrificing accuracy of the returned images. We adopt a group testing formulation and design the decoding architecture using either dictionary learning or eigendecomposition. The latter is a plausible option for small-to-medium sized problems with high-dimensional global image descriptors, whereas dictionary learning is applicable in large-scale scenarios. We evaluate our approach for global descriptors obtained from both SIFT and CNN features. Experiments with standard image search benchmarks, including the Yahoo100M dataset comprising 100 million images, show that our method gives comparable (and sometimes superior) accuracy compared to exhaustive search while requiring only 10% of the vector operations and memory. Moreover, for the same search complexity, our method gives significantly better accuracy compared to approaches based on dimensionality reduction or locality sensitive hashing.

We present a new approach for efficient approximate nearest neighbor (ANN) search in high dimensional spaces, extending the idea of Product Quantization. We propose a two level product and vector quantization tree that reduces the number of vector comparisons required during tree traversal. Our approach also includes a novel highly parallelizable re-ranking method for candidate vectors by efficiently reusing already computed intermediate values. Due to its small memory footprint during traversal the method lends itself to an efficient, parallel GPU implementation. This Product Quantization Tree approach significantly outperforms recent state of the art methods for high dimensional nearest neighbor queries on standard reference datasets. Ours is the first work that demonstrates GPU performance superior to CPU performance on high dimensional, large scale ANN problems in time-critical real-world applications, like loop-closing in videos.

In this paper, we present a new algorithm for finding all intersections of three quadrics. The proposed method is algebraic in nature and it is considerably more efficient than the Groebner basis and resultant-based solutions previously used in computer vision applications. We identify several computer vision problems that are formulated and solved as systems of three quadratic equations and for which our algorithm readily delivers considerably faster results. Also, we propose new formulations of three important vision problems: absolute camera pose with unknown focal length, generalized pose-and-scale, and hand-eye calibration with known translation. These new formulations allow our algorithm to significantly outperform the state-of-the-art in speed.

Existing billion-scale nearest neighbor search systems have mostly been compared on a single dataset of a billion of SIFT vectors, where systems based on the Inverted Multi-Index (IMI) have been performing very well, achieving state-of-the-art recall in several milliseconds. SIFT-like descriptors, however, are quickly being replaced with descriptors based on deep neural networks (DNN) that provide better performance for many computer vision tasks. In this paper, we introduce a new dataset of one billion descriptors based on DNNs and reveal the relative inefficiency of IMI-based indexing for such descriptors compared to SIFT data. We then introduce two new indexing structures, the Non-Orthogonal Inverted Multi-Index (NO-IMI) and the Generalized Non-Orthogonal Inverted Multi-Index (GNO-IMI). We show that due to additional flexibility, the new structures are able to adapt to DNN descriptor distribution in a better way. In particular, extensive experiments on the new dataset demonstrate that these data structures provide considerably better trade-off between the speed of retrieval and recall, given similar amount of memory, as compared to the standard Inverted Multi-Index.

We propose the first algorithm for non-rigid 2D-to-3D shape matching, where the input is a 2D query shape as well as a 3D target shape and the output is a continuous matching curve represented as a closed contour on the 3D shape. We cast the problem as finding the shortest circular path on the product 3-manifold of the two shapes. We prove that the optimal matching can be computed in polynomial time with a (worst-case) complexity of O(m*n^2*log(n)), where m and n denote the number of vertices on the 2D and the 3D shape respectively. Quantitative evaluation confirms that the method provides excellent results for sketch-based deformable 3D shape retrieval.

In the past year, convolutional neural networks have been shown to perform extremely well for stereo estimation. However, current architectures rely on siamese networks which exploit concatenation followed by further processing layers, requiring a minute of GPU computation per image pair. In contrast, in this paper we propose a matching network which is able to produce very accurate results in less than a second of GPU computation. Towards this goal, we exploit a product layer which simply computes the inner product between the two representations of a siamese architecture. We train our network by treating the problem as multi-class classification, where the classes are all possible disparities. This allows us to get calibrated scores, which result in much better matching performance when compared to existing approaches.

As a key component in many computer vision systems, optical flow estimation, especially with large displacements, remains an open problem. In this paper we present a simple but powerful matching method works in a coarse-to-fine scheme for optical flow estimation. Inspired by the nearest neighbor field (NNF) algorithms, our approach, called CPM (Coarse-to-fine PatchMatch), blends an efficient random search strategy with the coarse-to-fine scheme for optical flow problem. Unlike existing NNF techniques, which is efficient but the results is often too noisy for optical flow caused by the lack of global regularization, we propose a propagation step with constrained random search radius between adjacent levels on the hierarchical architecture. The resulting correspondences enjoys a built-in smoothing effect, which is more suited for optical flow estimation than NNF techniques. Furthermore, our approach can also capture the tiny structures with large motions which is a problem for traditional coarse-to-fine optical flow algorithms. Interpolated by an edge-preserving interpolation method (EpicFlow), our method outperforms the state of the art on MPI-Sintel and KITTI, and runs much faster than the competing methods.

We propose a method to recover the shape of a 3D room from a full-view indoor panorama. Our algorithm can automatically infer a 3D shape from a collection of partially oriented superpixel facets and line segments. The core part of the algorithm is a constraint graph, which includes lines and superpixels as vertices, and encodes their geometric relations as edges. A novel approach is proposed to perform 3D reconstruction based on the constraint graph by solving all the geometric constraints as constrained linear least-squares. The selected constraints used for reconstruction are identified using an occlusion detection method with a Markov random field. Experiments show that our method can recover room shapes that can not be addressed by previous approaches. Our method is also efficient, that is, the inference time for each panorama is less than 1 minute.