CVPR 2020 论文列表

We present a novel unsupervised feature representation learning method, Visual Commonsense Region-based Convolutional Neural Network (VC R-CNN), to serve as an improved visual region encoder for high-level tasks such as captioning and VQA. Given a set of detected object regions in an image (e.g., using Faster R-CNN), like any other unsupervised feature learning methods (e.g., word2vec), the proxy training objective of VC R-CNN is to predict the contextual objects of a region. However, they are fundamentally different: the prediction of VC R-CNN is by using causal intervention: P(Y|do(X)), while others are by using the conventional likelihood: P(Y|X). This is also the core reason why VC R-CNN can learn "sense-making" knowledge like chair can be sat -- while not just "common" co-occurrences such as chair is likely to exist if table is observed. We extensively apply VC R-CNN features in prevailing models of three popular tasks: Image Captioning, VQA, and VCR, and observe consistent performance boosts across them, achieving many new state-of-the-arts.

Vision-dialog navigation posed as a new holy-grail task in vision-language disciplinary targets at learning an agent endowed with the capability of constant conversation for help with natural language and navigating according to human responses. Besides the common challenges faced in visual language navigation, vision-dialog navigation also requires to handle well with the language intentions of a series of questions about the temporal context from dialogue history and co-reasoning both dialogs and visual scenes. In this paper, we propose the Cross-modal Memory Network (CMN) for remembering and understanding the rich information relevant to historical navigation actions. Our CMN consists of two memory modules, the language memory module (L-mem) and the visual memory module (V-mem). Specifically, L-mem learns latent relationships between the current language interaction and a dialog history by employing a multi-head attention mechanism. V-mem learns to associate the current visual views and the cross-modal memory about the previous navigation actions. The cross-modal memory is generated via a vision-to-language attention and a language-to-vision attention. Benefiting from the collaborative learning of the L-mem and the V-mem, our CMN is able to explore the memory about the decision making of historical navigation actions which is for the current step. Experiments on the CVDN dataset show that our CMN outperforms the previous state-of-the-art model by a significant margin on both seen and unseen environments.

We introduce a new task, Video-and-Language Inference, for joint multimodal understanding of video and text. Given a video clip with aligned subtitles as premise, paired with a natural language hypothesis based on the video content, a model needs to infer whether the hypothesis is entailed or contradicted by the given video clip. A new large-scale dataset, named Violin (VIdeO-and-Language INference), is introduced for this task, which consists of 95,322 video-hypothesis pairs from 15,887 video clips, spanning over 582 hours of video. These video clips contain rich content with diverse temporal dynamics, event shifts, and people interactions, collected from two sources: (i) popular TV shows, and (ii) movie clips from YouTube channels. In order to address our new multimodal inference task, a model is required to possess sophisticated reasoning skills, from surface-level grounding (e.g., identifying objects and characters in the video) to in-depth commonsense reasoning (e.g., inferring causal relations of events in the video). We present a detailed analysis of the dataset and an extensive evaluation over many strong baselines, providing valuable insights on the challenges of this new task.

We propose ViewAL, a novel active learning strategy for semantic segmentation that exploits viewpoint consistency in multi-view datasets. Our core idea is that inconsistencies in model predictions across viewpoints provide a very reliable measure of uncertainty and encourage the model to perform well irrespective of the viewpoint under which objects are observed. To incorporate this uncertainty measure, we introduce a new viewpoint entropy formulation, which is the basis of our active learning strategy. In addition, we propose uncertainty computations on a superpixel level, which exploits inherently localized signal in the segmentation task, directly lowering the annotation costs. This combination of viewpoint entropy and the use of superpixels allows to efficiently select samples that are highly informative for improving the network. We demonstrate that our proposed active learning strategy not only yields the best-performing models for the same amount of required labeled data, but also significantly reduces labeling effort. For instance, our method achieves 95% of maximum achievable network performance using only 7%, 17%, and 24% labeled data on SceneNet-RGBD, ScanNet, and Matterport3D, respectively. On these datasets, the best state-of-the-art method achieves the same performance with 14%, 27% and 33% labeled data. Finally, we demonstrate that labeling using superpixels yields the same quality of ground-truth compared to labeling whole images, but requires 25% less time.

View-based approach that recognizes 3D shape through its projected 2D images has achieved state-of-the-art results for 3D shape recognition. The major challenge for view-based approach is how to aggregate multi-view features to be a global shape descriptor. In this work, we propose a novel view-based Graph Convolutional Neural Network, dubbed as view-GCN, to recognize 3D shape based on graph representation of multiple views in flexible view configurations. We first construct view-graph with multiple views as graph nodes, then design a graph convolutional neural network over view-graph to hierarchically learn discriminative shape descriptor considering relations of multiple views. The view-GCN is a hierarchical network based on local and non-local graph convolution for feature transform, and selective view-sampling for graph coarsening. Extensive experiments on benchmark datasets show that view-GCN achieves state-of-the-art results for 3D shape classification and retrieval.

Event cameras are novel sensors that output brightness changes in the form of a stream of asynchronous "events" instead of intensity frames. They offer significant advantages with respect to conventional cameras: high dynamic range (HDR), high temporal resolution, and no motion blur. Recently, novel learning approaches operating on event data have achieved impressive results. Yet, these methods require a large amount of event data for training, which is hardly available due the novelty of event sensors in computer vision research. In this paper, we present a method that addresses these needs by converting any existing video dataset recorded with conventional cameras to synthetic event data. This unlocks the use of a virtually unlimited number of existing video datasets for training networks designed for real event data. We evaluate our method on two relevant vision tasks, i.e., object recognition and semantic segmentation, and show that models trained on synthetic events have several benefits: (i) they generalize well to real event data, even in scenarios where standard-camera images are blurry or overexposed, by inheriting the outstanding properties of event cameras; (ii) they can be used for fine-tuning on real data to improve over state-of-the-art for both classification and semantic segmentation.

Video super-resolution, which aims at producing a high-resolution video from its corresponding low-resolution version, has recently drawn increasing attention. In this work, we propose a novel method that can effectively incorporate temporal information in a hierarchical way. The input sequence is divided into several groups, with each one corresponding to a kind of frame rate. These groups provide complementary information to recover missing details in the reference frame, which is further integrated with an attention module and a deep intra-group fusion module. In addition, a fast spatial alignment is proposed to handle videos with large motion. Extensive results demonstrate the capability of the proposed model in handling videos with various motion. It achieves favorable performance against state-of-the-art methods on several benchmark datasets.

In self-supervised spatio-temporal representation learning, the temporal resolution and long-short term characteristics are not yet fully explored, which limits representation capabilities of learned models. In this paper, we propose a novel self-supervised method, referred to as video Playback Rate Perception (PRP), to learn spatio-temporal representation in a simple-yet-effective way. PRP roots in a dilated sampling strategy, which produces self-supervision signals about video playback rates for representation model learning. PRP is implemented with a feature encoder, a classification module, and a reconstructing decoder, to achieve spatio-temporal semantic retention in a collaborative discrimination-generation manner. The discriminative perception model follows a feature encoder to prefer perceiving low temporal resolution and long-term representation by classifying fast-forward rates. The generative perception model acts as a feature decoder to focus on comprehending high temporal resolution and short-term representation by introducing a motion-attention mechanism. PRP is applied on typical video target tasks including action recognition and video retrieval. Experiments show that PRP outperforms state-of-the-art self-supervised models with significant margins. Code is available at github.com/yuanyao366/PRP.

We explore the task of Video Object Grounding (VOG), which grounds objects in videos referred to in natural language descriptions. Previous methods apply image grounding based algorithms to address VOG, fail to explore the object relation information and suffer from limited generalization. Here, we investigate the role of object relations in VOG and propose a novel framework VOGNet to encode multi-modal object relations via self-attention with relative position encoding. To evaluate VOGNet, we propose novel contrasting sampling methods to generate more challenging grounding input samples, and construct a new dataset called ActivityNet-SRL (ASRL) based on existing caption and grounding datasets. Experiments on ASRL validate the need of encoding object relations in VOG, and our VOGNet outperforms competitive baselines by a significant margin.

Motion is a salient cue to recognize actions in video. Modern action recognition models leverage motion information either explicitly by using optical flow as input or implicitly by means of 3D convolutional filters that simultaneously capture appearance and motion information. This paper proposes an alternative approach based on a learnable correlation operator that can be used to establish frame-to-frame matches over convolutional feature maps in the different layers of the network. The proposed architecture enables the fusion of this explicit temporal matching information with traditional appearance cues captured by 2D convolution. Our correlation network compares favorably with widely-used 3D CNNs for video modeling, and achieves competitive results over the prominent two-stream network while being much faster to train. We empirically demonstrate that correlation networks produce strong results on a variety of video datasets, and outperform the state of the art on four popular benchmarks for action recognition: Kinetics, Something-Something, Diving48, and Sports1M.

We propose a modified variational autoencoder (VAE) architecture built on top of Mask R-CNN for instance-level video segmentation and tracking. The method builds a shared encoder and three parallel decoders, yielding three disjoint branches for predictions of future frames, object detection boxes, and instance segmentation masks. To effectively solve multiple learning tasks, we introduce a Gaussian Process model to enhance the statistical representation of VAE by relaxing the prior strong independent and identically distributed (iid) assumption of conventional VAEs and allowing potential correlations among extracted latent variables. The network learns embedded spatial interdependence and motion continuity in video data and creates a representation that is effective to produce high-quality segmentation masks and track multiple instances in diverse and unstructured videos. Evaluation on a variety of recently introduced datasets shows that our model outperforms previous methods and achieves the new best in class performance.

Behavior prediction in dynamic, multi-agent systems is an important problem in the context of self-driving cars, due to the complex representations and interactions of road components, including moving agents (e.g. pedestrians and vehicles) and road context information (e.g. lanes, traffic lights). This paper introduces VectorNet, a hierarchical graph neural network that first exploits the spatial locality of individual road components represented by vectors and then models the high-order interactions among all components. In contrast to most recent approaches, which render trajectories of moving agents and road context information as bird-eye images and encode them with convolutional neural networks (ConvNets), our approach operates on the primitive vector representation. By operating on the vectorized high definition (HD) maps and agent trajectories, we avoid lossy rendering and computationally intensive ConvNet encoding steps. To further boost VectorNet's capability in learning context features, we propose a novel auxiliary task to recover the randomly masked out map entities and agent trajectories based on their context. We evaluate VectorNet on our in-house behavior prediction benchmark and the recently released Argoverse forecasting dataset. Our method achieves on par or better performance than the competitive rendering approach on both benchmarks while saving over 70% of the model parameters with an order of magnitude reduction in FLOPs. It also obtains state-of-the-art performance on the Argoverse dataset.

Extracting road graphs from aerial images automatically is more efficient and costs less than from field acquisition. This can be done by a post-processing step that vectorizes road segmentation predicted by CNN, but imperfect predictions will result in road graphs with low connectivity. On the other hand, iterative next move exploration could construct road graphs with better road connectivity, but often focuses on local information and does not provide precise alignment with the real road. To enhance the road connectivity while maintaining the precise alignment between the graph and real road, we propose a point-based iterative graph exploration scheme with segmentation-cues guidance and flexible steps. In our approach, we represent the location of the next move as a 'point' that unifies the representation of multiple constraints such as the direction and step size in each moving step. Information cues such as road segmentation and road junctions are jointly detected and utilized to guide the next move and achieve better alignment of roads. We demonstrate that our proposed method has a considerable improvement over state-of-the-art road graph extraction methods in terms of F-measure and road connectivity metrics on common datasets.

The quality of images captured in bad weather is often affected by chromatic casts and low visibility due to the presence of atmospheric particles. Restoration of the color balance is often ignored in most of the existing image de-hazing methods. In this paper, we propose a varicolored end-to-end image de-hazing network which restores the color balance in a given varicolored hazy image and recovers the haze-free image. The proposed network comprises of 1) Haze color correction (HCC) module and 2) Visibility improvement (VI) module. The proposed HCC module provides required attention to each color channel and generates a color balanced hazy image. While the proposed VI module processes the color balanced hazy image through novel inception attention block to recover the haze-free image. We also propose a novel approach to generate a large-scale varicolored synthetic hazy image database. An ablation study has been carried out to demonstrate the effect of different factors on the performance of the proposed network for image de-hazing. Three benchmark synthetic datasets have been used for quantitative analysis of the proposed network. Visual results on a set of real-world hazy images captured in different weather conditions demonstrate the effectiveness of the proposed approach for varicolored image de-hazing.

Non-blind deblurring is an important problem encountered in many image restoration tasks. The focus of non-blind deblurring is on how to suppress noise magnification during deblurring. In practice, it often happens that the noise level of input image is unknown and varies among different images. This paper aims at developing a deep learning framework for deblurring images with unknown noise level. Based on the framework of variational expectation maximization (EM), an iterative noise-blind deblurring scheme is proposed which integrates the estimation of noise level and the quantification of image prior uncertainty. Then, the proposed scheme is unrolled to a neural network (NN) where image prior is modeled by NN with uncertainty quantification. Extensive experiments showed that the proposed method not only outperformed existing noise-blind deblurring methods by a large margin, but also outperformed those state-of-the-art image deblurring methods designed/trained with known noise level.

Context information is critical for image semantic segmentation. Especially in indoor scenes, the large variation of object scales makes spatial-context an important factor for improving the segmentation performance. Thus, in this paper, we propose a novel variational context-deformable (VCD) module to learn adaptive receptive-field in a structured fashion. Different from standard ConvNets, which share fixed-size spatial context for all pixels, the VCD module learns a deformable spatial-context with the guidance of depth information: depth information provides clues for identifying real local neighborhoods. Specifically, adaptive Gaussian kernels are learned with the guidance of multimodal information. By multiplying the learned Gaussian kernel with standard convolution filters, the VCD module can aggregate flexible spatial context for each pixel during convolution. The main contributions of this work are as follows: 1) a novel VCD module is proposed, which exploits learnable Gaussian kernels to enable feature learning with structured adaptive-context; 2) variational Bayesian probabilistic modeling is introduced for the training of VCD module, which can make it continuous and more stable; 3) a perspective-aware guidance module is designed to take advantage of multi-modal information for RGB-D segmentation. We evaluate the proposed approach on three widely-used datasets, and the performance improvement has shown the effectiveness of the proposed method.

Comprehensive visual understanding requires detection frameworks that can effectively learn and utilize object interactions while analyzing objects individually. This is the main objective in Human-Object Interaction (HOI) detection task. In particular, relative spatial reasoning and structural connections between objects are essential cues for analyzing interactions, which is addressed by the proposed Visual-Spatial-Graph Network (VSGNet) architecture. VSGNet extracts visual features from the human-object pairs, refines the features with spatial configurations of the pair, and utilizes the structural connections between the pair via graph convolutions. The performance of VSGNet is thoroughly evaluated using the Verbs in COCO (V-COCO) dataset. Experimental results indicate that VSGNet outperforms state-of-the-art solutions by 8% or 4 mAP.

Methods for teaching machines to answer visual questions have made significant progress in recent years, but current methods still lack important human capabilities, including integrating new visual classes and concepts in a modular manner, providing explanations for the answers and handling new domains without explicit examples. We propose a novel method that consists of two main parts: generating a question graph representation, and an answering procedure, guided by the abstract structure of the question graph to invoke an extendable set of visual estimators. Training is performed for the language part and the visual part on their own, but unlike existing schemes, the method does not require any training using images with associated questions and answers. This approach is able to handle novel domains (extended question types and new object classes, properties and relations) as long as corresponding visual estimators are available. In addition, it can provide explanations to its answers and suggest alternatives when questions are not grounded in the image. We demonstrate that this approach achieves both high performance and domain extensibility without any questions-answers training.

We present a novel single-view surface normal estimation method that combines traditional line and vanishing point analysis with a deep learning approach. Starting from a color image and a Manhattan line map, we use a deep neural network to regress on a dense normal map, and a dense Manhattan label map that identifies planar regions aligned with the Manhattan directions. We fuse the normal map and label map in a fully differentiable manner to produce a refined normal map as final output. To do so, we softly decompose the output into a Manhattan part and a non-Manhattan part. The Manhattan part is treated by discrete classification and vanishing points, while the non-Manhattan part is learned by direct supervision. Our method achieves state-of-the-art results on standard single-view normal estimation benchmarks. More importantly, we show that by using vanishing points and lines, our method has better generalization ability than existing works. In addition, we demonstrate how our surface normal network can improve the performance of depth estimation networks, both quantitatively and qualitatively, in particular, in 3D reconstructions of walls and other flat surfaces.