CVPR 2020 论文列表

Visual tempo characterizes the dynamics and the temporal scale of an action. Modeling such visual tempos of different actions facilitates their recognition. Previous works often capture the visual tempo through sampling raw videos at multiple rates and constructing an input-level frame pyramid, which usually requires a costly multi-branch network to handle. In this work we propose a generic Temporal Pyramid Network (TPN) at the feature-level, which can be flexibly integrated into 2D or 3D backbone networks in a plug-and-play manner. Two essential components of TPN, the source of features and the fusion of features, form a feature hierarchy for the backbone so that it can capture action instances at various tempos. TPN also shows consistent improvements over other challenging baselines on several action recognition datasets. Specifically, when equipped with TPN, the 3D ResNet-50 with dense sampling obtains a 2% gain on the validation set of Kinetics-400. A further analysis also reveals that TPN gains most of its improvements on action classes that have large variances in their visual tempos, validating the effectiveness of TPN.

Self-supervised audio-visual learning aims to capture useful representations of video by leveraging correspondences between visual and audio inputs. Existing approaches have focused primarily on matching semantic information between the sensory streams. We propose a novel self-supervised task to leverage an orthogonal principle: matching spatial information in the audio stream to the positions of sound sources in the visual stream. Our approach is simple yet effective. We train a model to determine whether the left and right audio channels have been flipped, forcing it to reason about spatial localization across the visual and audio streams. To train and evaluate our method, we introduce a large-scale video dataset, YouTube-ASMR-300K, with spatial audio comprising over 900 hours of footage. We demonstrate that understanding spatial correspondence enables models to perform better on three audio-visual tasks, achieving quantitative gains over supervised and self-supervised baselines that do not leverage spatial audio cues. We also show how to extend our self-supervised approach to 360 degree videos with ambisonic audio.

In this paper, we focus on weakly supervised learning with noisy training data for both classification and regression problems. We assume that the training outputs are collected from a mixture of a target and correlated noise distributions. Our proposed method simultaneously estimates the target distribution and the quality of each data which is defined as the correlation between the target and data generating distributions. The cornerstone of the proposed method is a Cholesky Block that enables modeling dependencies among mixture distributions in a differentiable manner where we maintain the distribution over the network weights. We first provide illustrative examples in both regression and classification tasks to show the effectiveness of the proposed method. Then, the proposed method is extensively evaluated in a number of experiments where we show that it constantly shows comparable or superior performances compared to existing baseline methods in the handling of noisy data.

In this work, we propose "tangent images," a spherical image representation that facilitates transferable and scalable 360 degree computer vision. Inspired by techniques in cartography and computer graphics, we render a spherical image to a set of distortion-mitigated, locally-planar image grids tangent to a subdivided icosahedron. By varying the resolution of these grids independently of the subdivision level, we can effectively represent high resolution spherical images while still benefiting from the low-distortion icosahedral spherical approximation. We show that training standard convolutional neural networks on tangent images compares favorably to the many specialized spherical convolutional kernels that have been developed, while also scaling efficiently to handle significantly higher spherical resolutions. Furthermore, because our approach does not require specialized kernels, we show that we can transfer networks trained on perspective images to spherical data without fine-tuning and with limited performance drop-off. Finally, we demonstrate that tangent images can be used to improve the quality of sparse feature detection on spherical images, illustrating its usefulness for traditional computer vision tasks like structure-from-motion and SLAM.

Co-salient object detection (CoSOD) is a newly emerging and rapidly growing branch of salient object detection (SOD), which aims to detect the co-occurring salient objects in multiple images. However, existing CoSOD datasets often have a serious data bias, which assumes that each group of images contains salient objects of similar visual appearances. This bias results in the ideal settings and the effectiveness of the models, trained on existing datasets, may be impaired in real-life situations, where the similarity is usually semantic or conceptual. To tackle this issue, we first collect a new high-quality dataset, named CoSOD3k, which contains 3,316 images divided into 160 groups with multiple level annotations, i.e., category, bounding box, object, and instance levels. CoSOD3k makes a significant leap in terms of diversity, difficulty and scalability, benefiting related vision tasks. Besides, we comprehensively summarize 34 cutting-edge algorithms, benchmarking 19 of them over four existing CoSOD datasets (MSRC, iCoSeg, Image Pair and CoSal2015) and our CoSOD3k with a total of 61K images (largest scale), and reporting group-level performance analysis. Finally, we discuss the challenge and future work of CoSOD. Our study would give a strong boost to growth in the CoSOD community. Benchmark toolbox and results are available on our project page.

In this paper, we present TailorNet, a neural model which predicts clothing deformation in 3D as a function of three factors: pose, shape and style (garment geometry), while retaining wrinkle detail. This goes beyond prior models, which are either specific to one style and shape, or generalize to different shapes producing smooth results, despite being style specific. Our hypothesis is that (even non-linear) combinations of examples smoothes out high frequency components such as fine-wrinkles, which makes learning the three factors jointly hard. At the heart of our technique is a decomposition of deformation into a high frequency and a low frequency component. While the low-frequency component is predicted from pose, shape and style parameters with an MLP, the high-frequency component is predicted with a mixture of shape-style specific pose models. The weights of the mixture are computed with a narrow bandwidth kernel to guarantee that only predictions with similar high-frequency patterns are combined. The style variation is obtained by computing, in a canonical pose, a subspace of deformation, which satisfies physical constraints such as inter-penetration, and draping on the body. TailorNet delivers 3D garments which retain the wrinkles from the physics based simulations (PBS) it is learned from, while running more than 1000 times faster. In contrast to classical PBS, TailorNet is easy to use and fully differentiable, which is crucial for computer vision and learning algorithms. Several experiments demonstrate TailorNet produces more realistic results than prior work, and even generates temporally coherent deformations on sequences of the AMASS dataset, despite being trained on static poses from a different dataset. To stimulate further research in this direction, we will make a dataset consisting of 55800 frames, as well as our model publicly available at https://virtualhumans.mpi-inf.mpg.de/tailornet/.

We present a method for solving two minimal problems for relative camera pose estimation from three views, which are based on three view correspondences of (i) three points and one line and (ii) three points and two lines through two of the points. These problems are too difficult to be efficiently solved by the state of the art Grobner basis methods. Our method is based on a new efficient homotopy continuation (HC) solver, which dramatically speeds up previous HC solving by specializing HC methods to generic cases of our problems. We show in simulated experiments that our solvers are numerically robust and stable under image noise. We show in real experiment that (i) SIFT features provide good enough point-and-line correspondences for three-view reconstruction and (ii) that we can solve difficult cases with too few or too noisy tentative matches where the state of the art structure from motion initialization fails.

Making accurate motion prediction of the surrounding traffic agents such as pedestrians, vehicles, and cyclists is crucial for autonomous driving. Recent data-driven motion prediction methods have attempted to learn to directly regress the exact future position or its distribution from massive amount of trajectory data. However, it remains difficult for these methods to provide multimodal predictions as well as integrate physical constraints such as traffic rules and movable areas. In this work we propose a novel two-stage motion prediction framework, Trajectory Proposal Network (TPNet). TPNet first generates a candidate set of future trajectories as hypothesis proposals, then makes the final predictions by classifying and refining the proposals which meets the physical constraints. By steering the proposal generation process, safe and multimodal predictions are realized. Thus this framework effectively mitigates the complexity of motion prediction problem while ensuring the multimodal output. Experiments on four large-scale trajectory prediction datasets, i.e. the ETH, UCY, Apollo and Argoverse datasets, show that TPNet achieves the state-of-the-art results both quantitatively and qualitatively.

We consider the problem of predicting the future trajectory of scene agents from egocentric views obtained from a moving platform. This problem is important in a variety of domains, particularly for autonomous systems making reactive or strategic decisions in navigation. In an attempt to address this problem, we introduce TITAN (Trajectory Inference using Targeted Action priors Network), a new model that incorporates prior positions, actions, and context to forecast future trajectory of agents and future ego-motion. In the absence of an appropriate dataset for this task, we created the TITAN dataset that consists of 700 labeled video-clips (with odometry) captured from a moving vehicle on highly interactive urban traffic scenes in Tokyo. Our dataset includes 50 labels including vehicle states and actions, pedestrian age groups, and targeted pedestrian action attributes that are organized hierarchically corresponding to atomic, simple/complex-contextual, transportive, and communicative actions. To evaluate our model, we conducted extensive experiments on the TITAN dataset, revealing significant performance improvement against baselines and state-of-the-art algorithms. We also report promising results from our Agent Importance Mechanism (AIM), a module which provides insight into assessment of perceived risk by calculating the relative influence of each agent on the future ego-trajectory. The dataset is available at https://usa.honda-ri.com/titan

Representation learning is a fundamental part of modern computer vision, where abstract representations of data are encoded as tensors optimized to solve problems like image segmentation and inpainting. Recently, self-attention in the form of Non-Local Block has emerged as a powerful technique to enrich features, by capturing complex interdependencies in feature tensors. However, standard self-attention approaches leverage only spatial relationships, drawing similarities between vectors and overlooking correlations between channels. In this paper, we introduce a new method, called Tensor Element Self-Attention (TESA) that generalizes such work to capture interdependencies along all dimensions of the tensor using matricization. An order R tensor produces R results, one for each dimension. The results are then fused to produce an enriched output which encapsulates similarity among tensor elements. Additionally, we analyze self-attention mathematically, providing new perspectives on how it adjusts the singular values of the input feature tensor. With these new insights, we present experimental results demonstrating how TESA can benefit diverse problems including classification and instance segmentation. By simply adding a TESA module to existing networks, we substantially improve competitive baselines and set new state-of-the-art results for image inpainting on Celeb and low light raw-to-rgb image translation on SID.

Temporal modeling is key for action recognition in videos. It normally considers both short-range motions and long-range aggregations. In this paper, we propose a Temporal Excitation and Aggregation (TEA) block, including a motion excitation (ME) module and a multiple temporal aggregation (MTA) module, specifically designed to capture both short- and long-range temporal evolution. In particular, for short-range motion modeling, the ME module calculates the feature-level temporal differences from spatiotemporal features. It then utilizes the differences to excite the motion-sensitive channels of the features. The long-range temporal aggregations in previous works are typically achieved by stacking a large number of local temporal convolutions. Each convolution processes a local temporal window at a time. In contrast, the MTA module proposes to deform the local convolution to a group of sub-convolutions, forming a hierarchical residual architecture. Without introducing additional parameters, the features will be processed with a series of sub-convolutions, and each frame could complete multiple temporal aggregations with neighborhoods. The final equivalent receptive field of temporal dimension is accordingly enlarged, which is capable of modeling the long-range temporal relationship over distant frames. The two components of the TEA block are complementary in temporal modeling. Finally, our approach achieves impressive results at low FLOPs on several action recognition benchmarks, such as Kinetics, Something-Something, HMDB51, and UCF101, which confirms its effectiveness and efficiency.

Video super-resolution (VSR) aims to restore a photo-realistic high-resolution (HR) video frame from both its corresponding low-resolution (LR) frame (reference frame) and multiple neighboring frames (supporting frames). Due to varying motion of cameras or objects, the reference frame and each support frame are not aligned. Therefore, temporal alignment is a challenging yet important problem for VSR. Previous VSR methods usually utilize optical flow between the reference frame and each supporting frame to warp the supporting frame for temporal alignment. However, both inaccurate flow and the image-level warping strategy will lead to artifacts in the warped supporting frames. To overcome the limitation, we propose a temporally-deformable alignment network (TDAN) to adaptively align the reference frame and each supporting frame at the feature level without computing optical flow. The TDAN uses features from both the reference frame and each supporting frame to dynamically predict offsets of sampling convolution kernels. By using the corresponding kernels, TDAN transforms supporting frames to align with the reference frame. To predict the HR video frame, a reconstruction network taking aligned frames and the reference frame is utilized. Experimental results demonstrate that the TDAN is capable of alleviating occlusions and artifacts for temporal alignment and the TDAN-based VSR model outperforms several recent state-of-the-art VSR networks with a comparable or even much smaller model size. The source code and pre-trained models are released in https://github.com/YapengTian/TDAN-VSR.

The state of the art person search methods separate person search into detection and re-ID stages, but ignore the consistency between these two stages. The general person detector has no special attention on the query target; The re-ID model is trained on hand-drawn bounding boxes which are not available in person search. To address the consistency problem, we introduce a Task-Consist Two-Stage (TCTS) person search framework, includes an identity-guided query (IDGQ) detector and a Detection Results Adapted (DRA) re-ID model. In the detection stage, the IDGQ detector learns an auxiliary identity branch to compute query similarity scores for proposals. With consideration of the query similarity scores and foreground score, IDGQ produces query-like bounding boxes for the re-ID stage. In the re-ID stage, we predict identity labels of detected bounding boxes, and use these examples to construct a more practical mixed train set for the DRA model. Training on the mixed train set improves the robustness of the re-ID stage to inaccurate detection. We evaluate our method on two benchmark datasets, CUHK-SYSU and PRW. Our framework achieves 93.9% of mAP and 95.1% of rank1 accuracy on CUHK-SYSU, outperforming the previous state of the art methods.

Security of modern Deep Neural Networks (DNNs) is under severe scrutiny as the deployment of these models become widespread in many intelligence-based applications. Most recently, DNNs are attacked through Trojan which can effectively infect the model during the training phase and get activated only through specific input patterns (i.e, trigger) during inference. In this work, for the first time, we propose a novel Targeted Bit Trojan(TBT) method, which can insert a targeted neural Trojan into a DNN through bit-flip attack. Our algorithm efficiently generates a trigger specifically designed to locate certain vulnerable bits of DNN weights stored in main memory (i.e., DRAM). The objective is that once the attacker flips these vulnerable bits, the network still operates with normal inference accuracy with benign input. However, when the attacker activates the trigger by embedding it with any input, the network is forced to classify all inputs to a certain target class. We demonstrate that flipping only several vulnerable bits identified by our method, using available bit-flip techniques (i.e, row-hammer), can transform a fully functional DNN model into a Trojan-infected model. We perform extensive experiments of CIFAR-10, SVHN and ImageNet datasets on both VGG-16 and Resnet-18 architectures. Our proposed TBT could classify 92 of test images to a target class with as little as 84 bit-flips out of 88 million weight bits on Resnet-18 for CIFAR10 dataset.

There are two main challenges in Visual Question Answering (VQA). The first one is that each model obtains its strengths and shortcomings when applied to several questions; what is more, the "ceiling effect" for specific questions is difficult to overcome with simple consecutive training. The second challenge is that even the state-of-the-art dataset is of large scale, questions targeted at a single image are off in format and lack diversity in content. We introduce our self-questioning model with multi-agent training: TA-student VQA. This framework differs from standard VQA algorithms by involving question-generating mechanisms and collaborative learning questions between question-answering agents. Thus, TA-student VQA overcomes the limitation of the content diversity and format variation of questions and improves the overall performance of multiple question-answering agents. We evaluate our model on VQA-v2, which outperforms algorithms without such mechanisms. In addition, TA-student VQA achieves a greater model capacity, allowing it to answer more generated questions in addition to those in the annotated datasets.

In this paper, we propose a data privacy-preserving and communication efficient distributed GAN learning framework named Distributed Asynchronized Discriminator GAN (AsynDGAN). Our proposed framework aims to train a central generator learns from distributed discriminator, and use the generated synthetic image solely to train the segmentation model. We validate the proposed framework on the application of health entities learning problem which is known to be privacy sensitive. Our experiments show that our approach: 1) could learn the real image's distribution from multiple datasets without sharing the patient's raw data. 2) is more efficient and requires lower bandwidth than other distributed deep learning methods. 3) achieves higher performance compared to the model trained by one real dataset, and almost the same performance compared to the model trained by all real datasets. 4) has provable guarantees that the generator could learn the distributed distribution in an all important fashion thus is unbiased.We release our AsynDGAN source code at: https://github.com/tommy-qichang/AsynDGAN

Existing methods on video captioning have made great efforts to identify objects/instances in videos, but few of them emphasize the prediction of action. As a result, the learned models are likely to depend heavily on the prior of training data, such as the co-occurrence of objects, which may cause an enormous divergence between the generated descriptions and the video content. In this paper, we explicitly emphasize the importance of action by predicting visually-related syntax components including subject, object and predicate. Specifically, we propose a Syntax-Aware Action Targeting (SAAT) module that firstly builds a self-attended scene representation to draw global dependence among multiple objects within a scene, and then decodes the visually-related syntax components by setting different queries. After targeting the action, indicated by predicate, our captioner learns an attention distribution over the predicate and the previously predicted words to guide the generation of the next word. Comprehensive experiments on MSVD and MSR-VTT datasets demonstrate the efficacy of the proposed model.

We introduce a new paradigm, `measure synchronization', for synchronizing graphs with measure-valued edges. We formulate this problem as maximization of the cycle-consistency in the space of probability measures over relative rotations. In particular, we aim at estimating marginal distributions of absolute orientations by synchronizing the `conditional' ones, which are defined on the Riemannian manifold of quaternions. Such graph optimization on distributions-on-manifolds enables a natural treatment of multimodal hypotheses, ambiguities and uncertainties arising in many computer vision applications such as SLAM, SfM, and object pose estimation. We first formally define the problem as a generalization of the classical rotation graph synchronization, where in our case the vertices denote probability measures over rotations. We then measure the quality of the synchronization by using Sinkhorn divergences, which reduces to other popular metrics such as Wasserstein distance or the maximum mean discrepancy as limit cases. We propose a nonparametric Riemannian particle optimization approach to solve the problem. Even though the problem is non-convex, by drawing a connection to the recently proposed sparse optimization methods, we show that the proposed algorithm converges to the global optimum in a special case of the problem under certain conditions. Our qualitative and quantitative experiments show the validity of our approach and we bring in new perspectives to the study of synchronization.

View synthesis allows for the generation of new views of a scene given one or more images. This is challenging; it requires comprehensively understanding the 3D scene from images. As a result, current methods typically use multiple images, train on ground-truth depth, or are limited to synthetic data. We propose a novel end-to-end model for this task using a single image at test time; it is trained on real images without any ground-truth 3D information. To this end, we introduce a novel differentiable point cloud renderer that is used to transform a latent 3D point cloud of features into the target view. The projected features are decoded by our refinement network to inpaint missing regions and generate a realistic output image. The 3D component inside of our generative model allows for interpretable manipulation of the latent feature space at test time, e.g. we can animate trajectories from a single image. Additionally, we can generate high resolution images and generalise to other input resolutions. We outperform baselines and prior work on the Matterport, Replica, and RealEstate10K datasets.