CVPR 2019 Papers

Omni-directional cameras have many advantages overconventional cameras in that they have a much wider field-of-view (FOV). Accordingly, several approaches have beenproposed recently to apply convolutional neural networks(CNNs) to omni-directional images for various visual tasks.However, most of them use image representations defined inthe Euclidean space after transforming the omni-directionalviews originally formed in the non-Euclidean space. Thistransformation leads to shape distortion due to nonuniformspatial resolving power and the loss of continuity. Theseeffects make existing convolution kernels experience diffi-culties in extracting meaningful information. This paper presents a novel method to resolve such prob-lems of applying CNNs to omni-directional images. Theproposed method utilizes a spherical polyhedron to rep-resent omni-directional views. This method minimizes thevariance of the spatial resolving power on the sphere sur-face, and includes new convolution and pooling methodsfor the proposed representation. The proposed method canalso be adopted by any existing CNN-based methods. Thefeasibility of the proposed method is demonstrated throughclassification, detection, and semantic segmentation taskswith synthetic and real datasets.

We propose sphere generative adversarial network (GAN), a novel integral probability metric (IPM)-based GAN. Sphere GAN uses the hypersphere to bound IPMs in the objective function. Thus, it can be trained stably. On the hypersphere, sphere GAN exploits the information of higher-order statistics of data using geometric moment matching, thereby providing more accurate results. In the paper, we mathematically prove the good properties of sphere GAN. In experiments, sphere GAN quantitatively and qualitatively surpasses recent state-of-the-art GANs for unsupervised image generation problems with the CIFAR-10, STL-10, and LSUN bedroom datasets. Source code is available at https://github.com/pswkiki/SphereGAN.

We propose a learning approach to corner detection for event-based cameras that is stable even under fast and abrupt motions. Event-based cameras offer high temporal resolution, power efficiency, and high dynamic range. However, the properties of event-based data are very different compared to standard intensity images, and simple extensions of corner detection methods designed for these images do not perform well on event-based data. We first introduce an efficient way to compute a time surface that is invariant to the speed of the objects. We then show that we can train a Random Forest to recognize events generated by a moving corner from our time surface. Random Forests are also extremely efficient, and therefore a good choice to deal with the high capture frequency of event-based cameras ---our implementation processes up to 1.6Mev/s on a single CPU. Thanks to our time surface formulation and this learning approach, our method is significantly more robust to abrupt changes of direction of the corners compared to previous ones. Our method also naturally assigns a confidence score for the corners, which can be useful for postprocessing. Moreover, we introduce a high-resolution dataset suitable for quantitative evaluation and comparison of corner detection methods for event-based cameras. We call our approach SILC, for Speed Invariant Learned Corners, and compare it to the state-of-the-art with extensive experiments, showing better performance.

How much can we infer about a person's looks from the way they speak? In this paper, we study the task of reconstructing a facial image of a person from a short audio recording of that person speaking. We design and train a deep neural network to perform this task using millions of natural Internet/Youtube videos of people speaking. During training, our model learns voice-face correlations that allow it to produce images that capture various physical attributes of the speakers such as age, gender and ethnicity. This is done in a self-supervised manner, by utilizing the natural co-occurrence of faces and speech in Internet videos, without the need to model attributes explicitly. We evaluate and numerically quantify how--and in what manner--our Speech2Face reconstructions, obtained directly from audio, resemble the true face images of the speakers.

Developing high light efficiency imaging techniques to retrieve high dimensional optical signal is a long-term goal in computational photography. Multispectral imaging, which captures images of different wavelengths and boosting the abilities for revealing scene properties, has developed rapidly in the last few decades. From scanning method to snapshot imaging, the limit of light collection efficiency is kept being pushed which enables wider applications especially under the light-starved scenes. In this work, we propose a novel multispectral imaging technique, that could capture the multispectral images with a high light efficiency. Through investigating the dispersive blur caused by spectral dispersers and introducing the difference of blur (DoB) constraints, we propose a basic theory for capturing multispectral information from a single dispersive-blurred image and an additional spectrum of an arbitrary point in the scene. Based on the theory, we design a prototype system and develop an optimization algorithm to realize snapshot multispectral imaging. The effectiveness of the proposed method is verified on both the synthetic data and real captured images.

In this paper, we propose a new measure to gauge the complexity of image classification problems. Given an annotated image dataset, our method computes a complexity measure called the cumulative spectral gradient (CSG) which strongly correlates with the test accuracy of convolutional neural networks (CNN). The CSG measure is derived from the probabilistic divergence between classes in a spectral clustering framework. We show that this metric correlates with the overall separability of the dataset and thus its inherent complexity. As will be shown, our metric can be used for dataset reduction, to assess which classes are more difficult to disentangle, and approximate the accuracy one could expect to get with a CNN. Results obtained on 11 datasets and three CNN models reveal that our method is more accurate and faster than previous complexity measures.

In this paper, we present a unified, end-to-end trainable spatiotemporal CNN model for VOS, which consists of two branches, i.e., the temporal coherence branch and the spatial segmentation branch. Specifically, the temporal coherence branch pretrained in an adversarial fashion from unlabeled video data, is designed to capture the dynamic appearance and motion cues of video sequences to guide object segmentation. The spatial segmentation branch focuses on segmenting objects accurately based on the learned appearance and motion cues. To obtain accurate segmentation results, we design a coarse-to-fine process to sequentially apply a designed attention module on multi-scale feature maps, and concatenate them to produce the final prediction. In this way, the spatial segmentation branch is enforced to gradually concentrate on object regions. These two branches are jointly fine-tuned on video segmentation sequences in an end-to-end manner. Several experiments are carried out on three challenging datasets (i.e., DAVIS-2016, DAVIS-2017 and Youtube-Object) to show that our method achieves favorable performance against the state-of-the-arts. Code is available at https://github.com/longyin880815/STCNN.

The need for efficiently finding the video content a user wants is increasing because of the erupting of user-generated videos on the Web. Existing keyword-based or content-based video retrieval methods usually determine what occurs in a video but not when and where. In this paper, we make an answer to the question of when and where by formulating a new task, namely spatio-temporal video re-localization. Specifically, given a query video and a reference video, spatio-temporal video re-localization aims to localize tubelets in the reference video such that the tubelets semantically correspond to the query. To accurately localize the desired tubelets in the reference video, we propose a novel warp LSTM network, which propagates the spatio-temporal information for a long period and thereby captures the corresponding long-term dependencies. Another issue for spatio-temporal video re-localization is the lack of properly labeled video datasets. Therefore, we reorganize the videos in the AVA dataset to form a new dataset for spatio-temporal video re-localization research. Extensive experimental results show that the proposed model achieves superior performances over the designed baselines on the spatio-temporal video re-localization task.

Automatic generation of video captions is a fundamental challenge in computer vision. Recent techniques typically employ a combination of Convolutional Neural Networks (CNNs) and Recursive Neural Networks (RNNs) for video captioning. These methods mainly focus on tailoring sequence learning through RNNs for better caption generation, whereas off-the-shelf visual features are borrowed from CNNs. We argue that careful designing of visual features for this task is equally important, and present a visual feature encoding technique to generate semantically rich captions using Gated Recurrent Units (GRUs). Our method embeds rich temporal dynamics in visual features by hierarchically applying Short Fourier Transform to CNN features of the whole video. It additionally derives high level semantics from an object detector to enrich the representation with spatial dynamics of the detected objects. The final representation is projected to a compact space and fed to a language model. By learning a relatively simple language model comprising two GRU layers, we establish new state-of-the-art on MSVD and MSR-VTT datasets for METEOR and ROUGE_L metrics.

Joint filtering mainly uses an additional guidance image as a prior and transfers its structures to the target image in the filtering process. Different from existing algorithms that rely on locally linear models or hand-designed objective functions to extract the structural information from the guidance image, we propose a new joint filter based on a spatially variant linear representation model (SVLRM), where the target image is linearly represented by the guidance image. However, the SVLRM leads to a highly ill-posed problem. To estimate the linear representation coefficients, we develop an effective algorithm based on a deep convolutional neural network (CNN). The proposed deep CNN (constrained by the SVLRM) is able to estimate the spatially variant linear representation coefficients which are able to model the structural information of both the guidance and input images. We show that the proposed algorithm can be effectively applied to a variety of applications, including depth/RGB image upsampling and restoration, flash/no-flash image deblurring, natural image denoising, scale-aware filtering, etc. Extensive experimental results demonstrate that the proposed algorithm performs favorably against state-of-the-art methods that have been specially designed for each task.

How to proper encode high-order object relation in the detection system without any external knowledge? How to leverage the information between co-occurrence and locations of objects for better reasoning? These questions are key challenges towards large-scale object detection system that aims to recognize thousands of objects entangled with complex spatial and semantic relationships nowadays. Distilling key relations that may affect object recognition is crucially important since treating each region separately leads to a big performance drop when facing heavy long-tail data distributions and plenty of confusing categories. Recent works try to encode relation by constructing graphs, e.g. using handcraft linguistic knowledge between classes or implicitly learning a fully-connected graph between regions. However, the handcraft linguistic knowledge cannot be individualized for each image due to the semantic gap between linguistic and visual context while the fully-connected graph is inefficient and noisy by incorporating redundant and distracted relations/edges from irrelevant objects and backgrounds. In this work, we introduce a Spatial-aware Graph Relation Network (SGRN) to adaptive discover and incorporate key semantic and spatial relationships for reasoning over each object. Our method considers the relative location layouts and interactions among which can be easily injected into any detection pipelines to boost the performance. Specifically, our SGRN integrates a graph learner module for learning a interpatable sparse graph structure to encode relevant contextual regions and a spatial graph reasoning module with learnable spatial Gaussian kernels to perform graph inference with spatial awareness. Extensive experiments verify the effectiveness of our method, e.g. achieving around 32% improvement on VG(3000 classes) and 28% on ADE in terms of mAP.

Recent advances in generative adversarial networks (GANs) have shown great potentials in realistic image synthesis whereas most existing works address synthesis realism in either appearance space or geometry space but few in both. This paper presents an innovative Spatial Fusion GAN (SF-GAN) that combines a geometry synthesizer and an appearance synthesizer to achieve synthesis realism in both geometry and appearance spaces. The geometry synthesizer learns contextual geometries of background images and transforms and places foreground objects into the background images unanimously. The appearance synthesizer adjust the color, brightness and styles of the foreground objects and embeds them into background images harmoniously, where a guided filter is incorporated for detail preserving. The two synthesizers are inter-connected as mutual references which can be trained end-to-end with little supervision. The SF-GAN has been evaluated in two tasks: (1) realistic scene text image synthesis for training better recognition models; (2) glass and hat wearing for realistic matching glasses and hats with real portraits. Qualitative and quantitative comparisons with the state-of-the-art demonstrate the superiority of the proposed SF-GAN.

Removing rain streaks from a single image has been drawing considerable attention as rain streaks can severely degrade the image quality and affect the performance of existing outdoor vision tasks. While recent CNN-based derainers have reported promising performances, deraining remains an open problem for two reasons. First, existing synthesized rain datasets have only limited realism, in terms of modeling real rain characteristics such as rain shape, direction and intensity. Second, there are no public benchmarks for quantitative comparisons on real rain images, which makes the current evaluation less objective. The core challenge is that real world rain/clean image pairs cannot be captured at the same time. In this paper, we address the single image rain removal problem in two ways. First, we propose a semi-automatic method that incorporates temporal priors and human supervision to generate a high-quality clean image from each input sequence of real rain images. Using this method, we construct a large-scale dataset of 29.5K rain/rain-free image pairs that covers a wide range of natural rain scenes. Second, to better cover the stochastic distribution of real rain streaks, we propose a novel SPatial Attentive Network (SPANet) to remove rain streaks in a local-to-global manner. Extensive experiments demonstrate that our network performs favorably against the state-of-the-art deraining methods.

Deep Neural Networks have achieved extraordinary results on image classification tasks, but have been shown to be vulnerable to attacks with carefully crafted perturbations of the input data. Although most attacks usually change values of many image's pixels, it has been shown that deep networks are also vulnerable to sparse alterations of the input. However, no computationally efficient method has been proposed to compute sparse perturbations. In this paper, we exploit the low mean curvature of the decision boundary, and propose SparseFool, a geometry inspired sparse attack that controls the sparsity of the perturbations. Extensive evaluations show that our approach computes sparse perturbations very fast, and scales efficiently to high dimensional data. We further analyze the transferability and the visual effects of the perturbations, and show the existence of shared semantic information across the images and the networks. Finally, we show that adversarial training can only slightly improve the robustness against sparse additive perturbations computed with SparseFool.

Ordinal regression attempts to solve classification problems in which categories are not independent, but rather follow a natural order. It is crucial to classify each class correctly while learning adequate interclass ordinal relationships. We present a simple and effective method that constrains these relationships among categories by seamlessly incorporating metric penalties into ground truth label representations. This encoding allows deep neural networks to automatically learn intraclass and interclass relationships without any explicit modification of the network architecture. Our method converts data labels into soft probability distributions that pair well with common categorical loss functions such as cross-entropy. We show that this approach is effective by using off-the-shelf classification and segmentation networks in four wildly different scenarios: image quality ranking, age estimation, horizon line regression, and monocular depth estimation. We demonstrate that our general-purpose method is very competitive with respect to specialized approaches, and adapts well to a variety of different network architectures and metrics.

As intelligent systems increasingly blend into our everyday life, artificial social intelligence becomes a prominent area of research. Intelligent systems must be socially intelligent in order to comprehend human intents and maintain a rich level of interaction with humans. Human language offers a unique unconstrained approach to probe through questions and reason through answers about social situations. This unconstrained approach extends previous attempts to model social intelligence through numeric supervision (e.g. sentiment and emotions labels). In this paper, we introduce Social-IQ, a unconstrained benchmark specifically designed to train and evaluate socially intelligent technologies. By providing a rich source of open-ended questions and answers, Social-IQ opens the door to explainable social intelligence. The dataset contains rigorously annotated and validated videos, questions and answers, as well as annotations for the complexity level of each question and answer. Social-IQ contains 1,250 natural in-the-wild social situations, 7,500 questions and 52,500 correct and incorrect answers. Although humans can reason about social situations with very high accuracy (95.08%), existing state-of-the-art computational models struggle on this task. As a result, Social-IQ brings novel challenges that will spark future research in social intelligence modeling, visual reasoning, and multimodal question answering (QA).

Discovering social relations, e.g., kinship, friendship, etc., from visual contents can make machines better interpret the behaviors and emotions of human beings. Existing studies mainly focus on recognizing social relations from still images while neglecting another important media--video. On one hand, the actions and storylines in videos provide more important cues for social relation recognition. On the other hand, the key persons may appear at arbitrary spatial-temporal locations, even not in one same image from beginning to the end. To overcome these challenges, we propose a Multi-scale Spatial-Temporal Reasoning (MSTR) framework to recognize social relations from videos. For the spatial representation, we not only adopt a temporal segment network to learn global action and scene information, but also design a Triple Graphs model to capture visual relations between persons and objects. For the temporal domain, we propose a Pyramid Graph Convolutional Network to perform temporal reasoning with multi-scale receptive fields, which can obtain both long-term and short-term storylines in videos. By this means, MSTR can comprehensively explore the multi-scale actions and storylines in spatial-temporal dimensions for social relation reasoning in videos. Extensive experiments on a new large-scale Video Social Relation dataset demonstrate the effectiveness of the proposed framework.

This paper addresses the problem of path prediction for multiple interacting agents in a scene, which is a crucial step for many autonomous platforms such as self-driving cars and social robots. We present SoPhie; an interpretable framework based on Generative Adversarial Network (GAN), which leverages two sources of information, the path history of all the agents in a scene, and the scene context information, using images of the scene. To predict a future path for an agent, both physical and social information must be leveraged. Previous work has not been successful to jointly model physical and social interactions. Our approach blends a social attention mechanism with physical attention that helps the model to learn where to look in a large scene and extract the most salient parts of the image relevant to the path. Whereas, the social attention component aggregates information across the different agent interactions and extracts the most important trajectory information from the surrounding neighbors. SoPhie also takes advantage of GAN to generates more realistic samples and to capture the uncertain nature of the future paths by modeling its distribution. All these mechanisms enable our approach to predict socially and physically plausible paths for the agents and to achieve state-of-the-art performance on several different trajectory forecasting benchmarks.

Several tasks in machine learning are evaluated using non-differentiable metrics such as mean average precision or Spearman correlation. However, their non-differentiability prevents from using them as objective functions in a learning framework. Surrogate and relaxation methods exist but tend to be specific to a given metric. In the present work, we introduce a new method to learn approximations of such non-differentiable objective functions. Our approach is based on a deep architecture that approximates the sorting of arbitrary sets of scores. It is trained virtually for free using synthetic data. This sorting deep (SoDeep) net can then be combined in a plug-and-play manner with existing deep architectures. We demonstrate the interest of our approach in three different tasks that require ranking: Cross-modal text-image retrieval, multi-label image classification and visual memorability ranking. Our approach yields very competitive results on these three tasks, which validates the merit and the flexibility of SoDeep as a proxy for sorting operation in ranking-based losses.