CVPR 2021 论文列表

Recent advances have shown that symmetry, a structural prior that most objects exhibit, can support a variety of single-view 3D understanding tasks. However, detecting 3D symmetry from an image remains a challenging task. Previous works either assume the symmetry is given or detect the symmetry with a heuristic-based method. In this paper, we present NeRD, a Neural 3D Reflection Symmetry Detector, which combines the strength of learning-based recognition and geometry-based reconstruction to accurately recover the normal direction of objects' mirror planes. Specifically, we enumerate the symmetry planes with a coarse-to-fine strategy and find the best ones by building 3D cost volumes to examine the intra-image pixel correspondence from the symmetry. Our experiments show that the symmetry planes detected with our method are significantly more accurate than the planes from direct CNN regression on both synthetic and real datasets. More importantly, we also demonstrate that the detected symmetry can be used to improve the performance of downstream tasks such as pose estimation and depth map regression by a wide margin over existing methods. The code of this paper has been made public at https://github.com/zhou13/nerd.

Generative Adversarial Networks (GANs) are currently an indispensable tool for visual editing, being a standard component of image-to-image translation and image restoration pipelines. Furthermore, GANs are especially useful for controllable generation since their latent spaces contain a wide range of interpretable directions, well suited for semantic editing operations. By gradually changing latent codes along these directions, one can produce impressive visual effects, unattainable without GANs. In this paper, we significantly expand the range of visual effects achievable with the state-of-the-art models, like StyleGAN2. In contrast to existing works, which mostly operate by latent codes, we discover interpretable directions in the space of the generator parameters. By several simple methods, we explore this space and demonstrate that it also contains a plethora of interpretable directions, which are an excellent source of non-trivial semantic manipulations. The discovered manipulations cannot be achieved by transforming the latent codes and can be used to edit both synthetic and real images. We release our code and models and hope they will serve as a handy tool for further efforts on GAN-based image editing.

We introduce two challenging datasets that reliably cause machine learning model performance to substantially degrade. The datasets are collected with a simple adversarial filtration technique to create datasets with limited spurious cues. Our datasets' real-world, unmodified examples transfer to various unseen models reliably, demonstrating that computer vision models have shared weaknesses. The first dataset is called ImageNet-A and is like the ImageNet test set, but it is far more challenging for existing models. We also curate an adversarial out-of-distribution detection dataset called ImageNet-O, which is the first out-of-distribution detection dataset created for ImageNet models. On ImageNet-A a DenseNet-121 obtains around 2% accuracy, an accuracy drop of approximately 90%, and its out-of-distribution detection performance on ImageNet-O is near random chance levels. We find that existing data augmentation techniques hardly boost performance, and using other public training datasets provides improvements that are limited. However, we find that improvements to computer vision architectures provide a promising path towards robust models.

With the increasing demand to efficiently deploy DNNs on mobile edge devices, it becomes much more important to reduce unnecessary computation and increase the execution speed. Prior methods towards this goal, including model compression and network architecture search (NAS), are largely performed independently, and do not fully consider compiler-level optimizations which is a must-do for mobile acceleration. In this work, we first propose (i) a general category of fine-grained structured pruning applicable to various DNN layers, and (ii) a comprehensive, compiler automatic code generation framework supporting different DNNs and different pruning schemes, which bridge the gap of model compression and NAS. We further propose NPAS, a compiler-aware unified network pruning and architecture search. To deal with large search space, we propose a meta-modeling procedure based on reinforcement learning with fast evaluation and Bayesian optimization, ensuring the total number of training epochs comparable with representative NAS frameworks. Our framework achieves 6.7ms, 5.9ms, and 3.9ms ImageNet inference times with 78.2%, 75% (MobileNet-V3 level), and 71% (MobileNet-V2 level) Top-1 accuracy respectively on an off-the-shelf mobile phone, consistently outperforming prior work.

We introduce NExT-QA, a rigorously designed video question answering (VideoQA) benchmark to advance video understanding from describing to explaining the temporal actions. Based on the dataset, we set up multi-choice and open-ended QA tasks targeting at causal action reasoning, temporal action reasoning and common scene comprehension. Through extensive analysis of baselines and established VideoQA techniques, we find that top-performing methods excel at shallow scene descriptions but are weak in causal and temporal action reasoning. Furthermore, the models that are effective on multi-choice QA, when adapted to open-ended QA, still struggle in generalizing the answers. This raises doubt on the ability of these models to reason and highlights possibilities for improvement. With detailed results for different question types and heuristic observations for future works, we hope NExT-QA will guide the next generation of VQA research to go beyond superficial description towards a deeper understanding of videos.

In this paper, we introduce NBNet, a novel framework for image denoising. Unlike previous works, we propose to tackle this challenging problem from a new perspective: noise reduction by image-adaptive projection. Specifically, we propose to train a network that can separate signal and noise by learning a set of reconstruction basis in the feature space. Subsequently, image denosing can be achieved by selecting corresponding basis of the signal subspace and projecting the input into such space. Our key insight is that projection can naturally maintain the local structure of input signal, especially for areas with low light or weak textures. Towards this end, we propose SSA, a non-local attention module we design to explicitly learn the basis generation as well as subspace projection. We further incorporate SSA with NBNet, a UNet structured network designed for end-to-end image denosing based. We conduct evaluations on benchmarks, including SIDD and DND, and NBNet achieves state-of-the-art performance on PSNR and SSIM with significantly less computational cost.

Automatically detecting/segmenting object(s) that blend in with their surroundings is difficult for current models. A major challenge is that the intrinsic similarities between such foreground objects and background surroundings make the features extracted by deep model indistinguishable. To overcome this challenge, an ideal model should be able to seek valuable, extra clues from the given scene and incorporate them into a joint learning framework for representation co-enhancement. With this inspiration, we design a novel Mutual Graph Learning (MGL) model, which generalizes the idea of conventional mutual learning from regular grids to the graph domain. Specifically, MGL decouples an image into two task-specific feature maps -- one for roughly locating the target and the other for accurately capturing its boundary details -- and fully exploits the mutual benefits by recurrently reasoning their high-order relations through graphs. Importantly, in contrast to most mutual learning approaches that use a shared function to model all between-task interactions, MGL is equipped with typed functions for handling different complementary relations to maximize information interactions. Experiments on challenging datasets, including CHAMELEON, CAMO and COD10K, demonstrate the effectiveness of our MGL with superior performance to existing state-of-the-art methods.

Graph-neural-networks (GNN) is a rising trend for few-shot learning. A critical component in GNN is the affinity. Typically, affinity in GNN is mainly computed in the feature space, e.g., pairwise features, and does not take fully advantage of semantic labels associated to these features. In this paper, we propose a novel Mutual CRF-GNN (MCGN). In this MCGN, the labels and features of support data are used by the CRF for inferring GNN affinities in a principled and probabilistic way. Specifically, we construct a Conditional Random Field (CRF) conditioned on labels and features of support data to infer a affinity in the label space. Such affinity is fed to the GNN as the node-wise affinity. GNN and CRF mutually contributes to each other in MCGN. For GNN, CRF provides valuable affinity information. For CRF, GNN provides better features for inferring affinity. Experimental results show that our approach outperforms state-of-the-arts on datasets miniImageNet, tieredImageNet, and CIFAR-FS on both 5-way 1-shot and 5-way 5-shot settings.

Multispectral photometric stereo (MPS) aims at recovering the surface normal of a scene from a single-shot multispectral image, which is known as an ill-posed problem. To make the problem well-posed, existing MPS methods rely on restrictive assumptions, such as shape prior, surfaces having a monochromatic with uniform albedo. This paper alleviates the restrictive assumptions in existing methods. We show that the problem becomes well-posed for a surface with a uniform chromaticity but spatially-varying albedos based on our new formulation. Specifically, if at least three (or two) scene points share the same chromaticity, the proposed method uniquely recovers their surface normals and spectral reflectance with the illumination of more than or equal to four (or five) spectral lights. Besides, our method can be made robust by having many (i.e., 4 or more) spectral bands using robust estimation techniques for conventional photometric stereo. Experiments on both synthetic and real-world scenes demonstrate the effectiveness of our method. Our data and result can be found at https://github.com/GH-HOME/MultispectralPS.git.

Unsupervised representation learning has proved to be a critical component of anomaly detection/localization in images. The challenges to learn such a representation are two-fold. Firstly, the sample size is not often large enough to learn a rich generalizable representation through conventional techniques. Secondly, while only normal samples are available at training, the learned features should be discriminative of normal and anomalous samples. Here, we propose to use the "distillation" of features at various layers of an expert network, which is pre-trained on ImageNet, into a simpler cloner network to tackle both issues. We detect and localize anomalies using the discrepancy between the expert and cloner networks' intermediate activation values given an input sample. We show that considering multiple intermediate hints in distillation leads to better exploitation of the expert's knowledge and a more distinctive discrepancy between the two networks, compared to utilizing only the last layer activation values. Notably, previous methods either fail in precise anomaly localization or need expensive region-based training. In contrast, with no need for any special or intensive training procedure, we incorporate interpretability algorithms in our novel framework to localize anomalous regions. Despite the striking difference between some test datasets and ImageNet, we achieve competitive or significantly superior results compared to SOTA on MNIST, F-MNIST, CIFAR-10, MVTecAD, Retinal-OCT, and two other medical datasets on both anomaly detection and localization.

Recent works have shown that convolutional networks have substantially improved the performance of multiple object tracking by simultaneously learning detection and appearance features. However, due to the local perception of the convolutional network structure itself, the long-range dependencies in both the spatial and temporal cannot be obtained efficiently. To incorporate the spatial layout, we propose to exploit the local correlation module to model the topological relationship between targets and their surrounding environment, which can enhance the discriminative power of our model in crowded scenes. Specifically, we establish dense correspondences of each spatial location and its context, and explicitly constrain the correlation volumes through self-supervised learning. To exploit the temporal context, existing approaches generally utilize two or more adjacent frames to construct an enhanced feature representation, but the dynamic motion scene is inherently difficult to depict via CNNs. Instead, our paper proposes a learnable correlation operator to establish frame-to-frame matches over convolutional feature maps in the different layers to align and propagate temporal context. With extensive experimental results on the MOT datasets, our approach demonstrates the effectiveness of correlation learning with the superior performance and obtains state-of-the-art MOTA of 76.5% and IDF1 of 73.6% on MOT17.

We present ARCH, a computational pathology (CP) multiple instance captioning dataset to facilitate dense supervision of CP tasks. Existing CP datasets focus on narrow tasks; ARCH on the other hand contains dense diagnostic and morphological descriptions for a range of stains, tissue types and pathologies. Using intrinsic dimensionality estimation, we show that ARCH is the only CP dataset to (ARCH-)rival its computer vision analog MS-COCO Captions. We conjecture that an encoder pre-trained on dense image captions learns transferable representations for most CP tasks. We support the conjecture with evidence that ARCH representation transfers to a variety of pathology sub-tasks better than ImageNet features or representations obtained via self-supervised or multi-task learning on pathology images alone. We release our best model and invite other researchers to test it on their CP tasks.

Despite the substantial progress of active learning for image recognition, there still lacks an instance-level active learning method specified for object detection. In this paper, we propose Multiple Instance Active Object Detection (MI-AOD), to select the most informative images for detector training by observing instance-level uncertainty. MI-AOD defines an instance uncertainty learning module, which leverages the discrepancy of two adversarial instance classifiers trained on the labeled set to predict instance uncertainty of the unlabeled set. MI-AOD treats unlabeled images as instance bags and feature anchors in images as instances, and estimates the image uncertainty by re-weighting instances in a multiple instance learning (MIL) fashion. Iterative instance uncertainty learning and re-weighting facilitate suppressing noisy instances, toward bridging the gap between instance uncertainty and image-level uncertainty. Experiments validate that MI-AOD sets a solid baseline for instance-level active learning. On commonly used object detection datasets, MI-AOD outperforms state-of-the-art methods with significant margins, particularly when the labeled sets are small.

Predicting multiple plausible future trajectories of the nearby vehicles is crucial for the safety of autonomous driving. Recent motion prediction approaches attempt to achieve such multimodal motion prediction by implicitly regularizing the feature or explicitly generating multiple candidate proposals. However, it remains challenging since the latent features may concentrate on the most frequent mode of the data while the proposal-based methods depend largely on the prior knowledge to generate and select the proposals. In this work, we propose a novel transformer framework for multimodal motion prediction, termed as mmTransformer. A novel network architecture based on stacked transformers is designed to model the multimodality at feature level with a set of fixed independent proposals. A region-based training strategy is then developed to induce the multimodality of the generated proposals. Experiments on Argoverse dataset show that the proposed model achieves the state-of-the-art performance on motion prediction, substantially improving the diversity and the accuracy of the predicted trajectories. Demo video and code are available at https://decisionforce. github.io/mmTransformer.

We develop an approach to learning visual representations that embraces multimodal data, driven by a combination of intra- and inter-modal similarity preservation objectives. Unlike existing visual pre-training methods, which solve a proxy prediction task in a single domain, our method exploits intrinsic data properties within each modality and semantic information from cross-modal correlation simultaneously, hence improving the quality of learned visual representations. By including multimodal training in a unified framework with different types of contrastive losses, our method can learn more powerful and generic visual features. We first train our model on COCO and evaluate the learned visual representations on various downstream tasks including image classification, object detection, and instance segmentation. For example, the visual representations pre-trained on COCO by our method achieve state-of-the-art top-1 validation accuracy of 55.3% on ImageNet classification, under the common transfer protocol. We also evaluate our method on the large-scale Stock images dataset and show its effectiveness on multi-label image tagging, and cross-modal retrieval tasks.

We address the problem of recovering multiple structures of different classes in a dataset contaminated by noise and outliers. In particular, we consider geometric structures defined by a mixture of underlying parametric models (e.g. planes and cylinders, homographies and fundamental matrices), and we tackle the robust fitting problem by preference analysis and clustering. We present a new algorithm, termed MultiLink, that simultaneously deals with multiple classes of models. MultiLink wisely combines on-the-fly model fitting and model selection in a novel linkage scheme that determines whether two clusters are to be merged. The resulting method features many practical advantages with respect to methods based on preference analysis, being faster, less sensitive to the inlier threshold, and able to compensate limitations deriving from hypotheses sampling. Experiments on several public datasets demonstrate that MultiLink favorably compares with state of the art alternatives, both in multi-class and single-class problems. Code is publicly made available for download.

We present MultiBodySync, a novel, end-to-end trainable multi-body motion segmentation and rigid registration framework for multiple input 3D point clouds. The two non-trivial challenges posed by this multi-scan multibody setting that we investigate are: (i) guaranteeing correspondence and segmentation consistency across multiple input point clouds capturing different spatial arrangements of bodies or body parts; and (ii) obtaining robust motion-based rigid body segmentation applicable to novel object categories. We propose an approach to address these issues that incorporates spectral synchronization into an iterative deep declarative network, so as to simultaneously recover consistent correspondences as well as motion segmentation. At the same time, by explicitly disentangling the correspondence and motion segmentation estimation modules, we achieve strong generalizability across different object categories. Our extensive evaluations demonstrate that our method is effective on various datasets ranging from rigid parts in articulated objects to individually moving objects in a 3D scene, be it single-view or full point clouds.

We present a novel method for multi-view depth estimation from a single video, which is a critical task in various applications, such as perception, reconstruction and robot navigation. Although previous learning-based methods have demonstrated compelling results, most works estimate depth maps of individual video frames independently, without taking into consideration the strong geometric and temporal coherence among the frames. Moreover, current state-of-the-art (SOTA) models mostly adopt a fully 3D convolution network for cost regularization and therefore require high computational cost, thus limiting their deployment in real-world applications. Our method achieves temporally coherent depth estimation results by using a novel Epipolar Spatio-Temporal (EST) transformer to explicitly associate geometric and temporal correlation with multiple estimated depth maps. Furthermore, to reduce the computational cost, inspired by recent Mixture-of-Experts models, we design a compact hybrid network consisting of a 2D context-aware network and a 3D matching network which learn 2D context information and 3D disparity cues separately. Extensive experiments demonstrate that our method achieves higher accuracy in depth estimation and significant speedup than the SOTA methods.

Existing approaches for multi-view multi-person 3D pose estimation explicitly establish cross-view correspondences to group 2D pose detections from multiple camera views and solve for the 3D pose estimation for each person. Establishing cross-view correspondences is challenging in multi-person scenes, and incorrect correspondences will lead to sub-optimal performance for the multi-stage pipeline. In this work, we present our multi-view 3D pose estimation approach based on plane sweep stereo to jointly address the cross-view fusion and 3D pose reconstruction in a single shot. Specifically, we propose to perform depth regression for each joint of each 2D pose in a target camera view. Cross-view consistency constraints are implicitly enforced by multiple reference camera views via the plane sweep algorithm to facilitate accurate depth regression. We adopt a coarse-to-fine scheme to first regress the person-level depth followed by a per-person joint-level relative depth estimation. 3D poses are obtained from a simple back-projection given the estimated depths. We evaluate our approach on benchmark datasets where it outperforms previous state-of-the-arts while being remarkably efficient.