CVPR 2020 论文列表

Existing deep learning methods for image deblurring typically train models using pairs of sharp images and their blurred counterparts. However, synthetically blurring images does not necessarily model the blurring process in real-world scenarios with sufficient accuracy. To address this problem, we propose a new method which combines two GAN models, i.e., a learning-to-Blur GAN (BGAN) and learning-to-DeBlur GAN (DBGAN), in order to learn a better model for image deblurring by primarily learning how to blur images. The first model, BGAN, learns how to blur sharp images with unpaired sharp and blurry image sets, and then guides the second model, DBGAN, to learn how to correctly deblur such images. In order to reduce the discrepancy between real blur and synthesized blur, a relativistic blur loss is leveraged. As an additional contribution, this paper also introduces a Real-World Blurred Image (RWBI) dataset including diverse blurry images. Our experiments show that the proposed method achieves consistently superior quantitative performance as well as higher perceptual quality on both the newly proposed dataset and the public GOPRO dataset.

Choosing a deep neural network architecture is a fundamental problem in applications that require balancing performance and parameter efficiency. Standard approaches rely on ad-hoc engineering or computationally expensive validation on a specific dataset. We instead attempt to quantify networks by their intrinsic capacity for unique and robust representations, enabling efficient architecture comparisons without requiring any data. Building upon theoretical connections between deep learning and sparse approximation, we propose the deep frame potential: a measure of coherence that is approximately related to representation stability but has minimizers that depend only on network structure. This provides a framework for jointly quantifying the contributions of architectural hyper-parameters such as depth, width, and skip connections. We validate its use as a criterion for model selection and demonstrate correlation with generalization error on a variety of common residual and densely connected network architectures.

Machine learning models are vulnerable to adversarial examples. For the black-box setting, current substitute attacks need pre-trained models to generate adversarial examples. However, pre-trained models are hard to obtain in real-world tasks. In this paper, we propose a data-free substitute training method (DaST) to obtain substitute models for adversarial black-box attacks without the requirement of any real data. To achieve this, DaST utilizes specially designed generative adversarial networks (GANs) to train the substitute models. In particular, we design a multi-branch architecture and label-control loss for the generative model to deal with the uneven distribution of synthetic samples. The substitute model is then trained by the synthetic samples generated by the generative model, which are labeled by the attacked model subsequently. The experiments demonstrate the substitute models produced by DaST can achieve competitive performance compared with the baseline models which are trained by the same train set with attacked models. Additionally, to evaluate the practicability of the proposed method on the real-world task, we attack an online machine learning model on the Microsoft Azure platform. The remote model misclassifies 98.35% of the adversarial examples crafted by our method. To the best of our knowledge, we are the first to train a substitute model for adversarial attacks without any real data.

Recent works imply that the channel pruning can be regarded as searching optimal sub-structure from unpruned networks. However, existing works based on this observation require training and evaluating a large number of structures, which limits their application. In this paper, we propose a novel differentiable method for channel pruning, named Differentiable Markov Channel Pruning (DMCP), to efficiently search the optimal sub-structure. Our method is differentiable and can be directly optimized by gradient descent with respect to standard task loss and budget regularization (e.g. FLOPs constraint). In DMCP, we model the channel pruning as a Markov process, in which each state represents for retaining the corresponding channel during pruning, and transitions between states denote the pruning process. In the end, our method is able to implicitly select the proper number of channels in each layer by the Markov process with optimized transitions. To validate the effectiveness of our method, we perform extensive experiments on Imagenet with ResNet and MobilenetV2. Results show our method can achieve consistent improvement than state-of-the-art pruning methods in various FLOPs settings.

Exploring the intrinsic interconnections between the knowledge encoded in PRe-trained Deep Neural Networks (PR-DNNs) of heterogeneous tasks sheds light on their mutual transferability, and consequently enables knowledge transfer from one task to another so as to reduce the training effort of the latter. In this paper, we propose the DEeP Attribution gRAph (DEPARA) to investigate the transferability of knowledge learned from PR-DNNs. In DEPARA, nodes correspond to the inputs and are represented by their vectorized attribution maps with regards to the outputs of the PR-DNN. Edges denote the relatedness between inputs and are measured by the similarity of their features extracted from the PR-DNN. The knowledge transferability of two PR-DNNs is measured by the similarity of their corresponding DEPARAs. We apply DEPARA to two important yet under-studied problems in transfer learning: pre-trained model selection and layer selection. Extensive experiments are conducted to demonstrate the effectiveness and superiority of the proposed method in solving both these problems. Code, data and models reproducing the results in this paper are available at https://github.com/zju-vipa/DEPARA.

Close-up talking heads are among the most common and salient object in video contents, such as face-to-face conversations in social media, teleconferences, news broadcasting, talk shows, etc. Due to the high sensitivity of human visual system to faces, compression distortions in talking heads videos are highly visible and annoying. To address this problem, we present a novel deep convolutional neural network (DCNN) method for very low bit rate video reconstruction of talking heads. The key innovation is a new DCNN architecture that can exploit the audio-video correlations to repair compression defects in the face region. We further improve reconstruction quality by embedding into our DCNN the encoder information of the video compression standards and introducing a constraining projection module in the network. Extensive experiments demonstrate that the proposed DCNN method outperforms the existing state-of-the-art methods on videos of talking heads.

We propose D3VO as a novel framework for monocular visual odometry that exploits deep networks on three levels -- deep depth, pose and uncertainty estimation. We first propose a novel self-supervised monocular depth estimation network trained on stereo videos without any external supervision. In particular, it aligns the training image pairs into similar lighting condition with predictive brightness transformation parameters. Besides, we model the photometric uncertainties of pixels on the input images, which improves the depth estimation accuracy and provides a learned weighting function for the photometric residuals in direct (feature-less) visual odometry. Evaluation results show that the proposed network outperforms state-of-the-art self-supervised depth estimation networks. D3VO tightly incorporates the predicted depth, pose and uncertainty into a direct visual odometry method to boost both the front-end tracking as well as the back-end non-linear optimization. We evaluate D3VO in terms of monocular visual odometry on both the KITTI odometry benchmark and the EuRoC MAV dataset. The results show that D3VO outperforms state-of-the-art traditional monocular VO methods by a large margin. It also achieves comparable results to state-of-the-art stereo/LiDAR odometry on KITTI and to the state-of-the-art visual-inertial odometry on EuRoC MAV, while using only a single camera.

A successful point cloud registration often lies on robust establishment of sparse matches through discriminative 3D local features. Despite the fast evolution of learning-based 3D feature descriptors, little attention has been drawn to the learning of 3D feature detectors, even less for a joint learning of the two tasks. In this paper, we leverage a 3D fully convolutional network for 3D point clouds, and propose a novel and practical learning mechanism that densely predicts both a detection score and a description feature for each 3D point. In particular, we propose a keypoint selection strategy that overcomes the inherent density variations of 3D point clouds, and further propose a self-supervised detector loss guided by the on-the-fly feature matching results during training. Finally, our method achieves state-of-the-art results in both indoor and outdoor scenarios, evaluated on 3DMatch and KITTI datasets, and shows its strong generalization ability on the ETH dataset. Towards practical use, we show that by adopting a reliable feature detector, sampling a smaller number of features is sufficient to achieve accurate and fast point cloud alignment.

The availability of large-scale datasets has helped unleash the true potential of deep convolutional neural networks (CNNs). However, for the single-image denoising problem, capturing a real dataset is an unacceptably expensive and cumbersome procedure. Consequently, image denoising algorithms are mostly developed and evaluated on synthetic data that is usually generated with a widespread assumption of additive white Gaussian noise (AWGN). While the CNNs achieve impressive results on these synthetic datasets, they do not perform well when applied on real camera images, as reported in recent benchmark datasets. This is mainly because the AWGN is not adequate for modeling the real camera noise which is signal-dependent and heavily transformed by the camera imaging pipeline. In this paper, we present a framework that models camera imaging pipeline in forward and reverse directions. It allows us to produce any number of realistic image pairs for denoising both in RAW and sRGB spaces. By training a new image denoising network on realistic synthetic data, we achieve the state-of-the-art performance on real camera benchmark datasets. The parameters in our models are 5 times lesser than the previous best method for RAW denoising. Furthermore, we demonstrate that the proposed framework generalizes beyond image denoising problem e.g., for color matching in stereoscopic cinema. The source code and pre-trained models are available at https://github.com/swz30/CycleISP.

Any solid object can be decomposed into a collection of convex polytopes (in short, convexes). When a small number of convexes are used, such a decomposition can be thought of as a piece-wise approximation of the geometry. This decomposition is fundamental in computer graphics, where it provides one of the most common ways to approximate geometry, for example, in real-time physics simulation. A convex object also has the property of being simultaneously an explicit and implicit representation: one can interpret it explicitly as a mesh derived by computing the vertices of a convex hull, or implicitly as the collection of half-space constraints or support functions. Their implicit representation makes them particularly well suited for neural network training, as they abstract away from the topology of the geometry they need to represent. However, at testing time, convexes can also generate explicit representations - polygonal meshes - which can then be used in any downstream application. We introduce a network architecture to represent a low dimensional family of convexes. This family is automatically derived via an auto-encoding process. We investigate the applications of this architecture including automatic convex decomposition, image to 3D reconstruction, and part-based shape retrieval.

Mammogram mass detection is of great clinical significance due to its high proportion in breast cancers. The information from cross views (i.e., mediolateral oblique and cranio-caudal) is highly related and complementary, and is helpful to make comprehensive decisions. However, unlike radiologists who are able to recognize masses with reasoning ability in cross-view images, most existing methods lack the ability to reason under the guidance of domain knowledge, thus it limits the performance. In this paper, we introduce bipartite graph convolutional network to endow existing methods with cross-view reasoning ability of radiologists in mammogram mass detection. The bipartite node sets are constructed by cross-view images respectively to represent relatively consistent regions in breasts, while the bipartite edge learns to model both inherent cross-view geometric constraints and appearance similarities between correspondences. Based on the bipartite graph, the information propagates methodically through correspondences and enables spatial visual features equipped with customized cross-view reasoning ability. Experimental results on DDSM dataset demonstrate that the proposed algorithm achieves state-of-the-art performance. Besides, visual analysis shows the model has a clear physical meaning, which is helpful for radiologists in clinical interpretation.

We present an approach for estimating surface normals from in-the-wild color images of faces. While data-driven strategies have been proposed for single face images, limited available ground truth data makes this problem difficult. To alleviate this issue, we propose a method that can leverage all available image and normal data, whether paired or not, thanks to a novel cross-modal learning architecture. In particular, we enable additional training with single modality data, either color or normal, by using two encoder-decoder networks with a shared latent space. The proposed architecture also enables face details to be transferred between the image and normal domains, given paired data, through skip connections between the image encoder and normal decoder. Core to our approach is a novel module that we call deactivable skip connections, which allows integrating both the auto-encoded and image-to-normal branches within the same architecture that can be trained end-to-end. This allows learning of a rich latent space that can accurately capture the normal information. We compare against state-of-the-art methods and show that our approach can achieve significant improvements, both quantitative and qualitative, with natural face images.

Applying the knowledge of an object detector trained on a specific domain directly onto a new domain is risky, as the gap between two domains can severely degrade model's performance. Furthermore, since different instances commonly embody distinct modal information in object detection scenario, the feature alignment of source and target domain is hard to be realized. To mitigate these problems, we propose a Graph-induced Prototype Alignment (GPA) framework to seek for category-level domain alignment via elaborate prototype representations. In the nutshell, more precise instance-level features are obtained through graph-based information propagation among region proposals, and, on such basis, the prototype representation of each class is derived for category-level domain alignment. In addition, in order to alleviate the negative effect of class-imbalance on domain adaptation, we design a Class-reweighted Contrastive Loss to harmonize the adaptation training process. Combining with Faster R-CNN, the proposed framework conducts feature alignment in a two-stage manner. Comprehensive results on various cross-domain detection tasks demonstrate that our approach outperforms existing methods with a remarkable margin. Our code is available at https://github.com/ChrisAllenMing/GPA-detection.

We present a general framework for exemplar-based image translation, which synthesizes a photo-realistic image from the input in a distinct domain (e.g., semantic segmentation mask, or edge map, or pose keypoints), given an exemplar image. The output has the style (e.g., color, texture) in consistency with the semantically corresponding objects in the exemplar. We propose to jointly learn the cross-domain correspondence and the image translation, where both tasks facilitate each other and thus can be learned with weak supervision. The images from distinct domains are first aligned to an intermediate domain where dense correspondence is established. Then, the network synthesizes images based on the appearance of semantically corresponding patches in the exemplar. We demonstrate the effectiveness of our approach in several image translation tasks. Our method is superior to state-of-the-art methods in terms of image quality significantly, with the image style faithful to the exemplar with semantic consistency. Moreover, we show the utility of our method for several applications.

Mining informative negative instances are of central importance to deep metric learning (DML). However, the hard-mining ability of existing DML methods is intrinsically limited by mini-batch training, where only a mini-batch of instances are accessible at each iteration. In this paper, we identify a "slow drift" phenomena by observing that the embedding features drift exceptionally slow even as the model parameters are updating throughout the training process. It suggests that the features of instances computed at preceding iterations can considerably approximate to their features extracted by current model. We propose a cross-batch memory (XBM) mechanism that memorizes the embeddings of past iterations, allowing the model to collect sufficient hard negative pairs across multiple mini-batches - even over the whole dataset. Our XBM can be directly integrated into general pair-based DML framework.We demonstrate that, without bells and whistles, XBM augmented DML can boost the performance considerably on image retrieval. In particular, with XBM, a simple contrastive loss can have large R@1 improvements of 12%-22.5% on three large-scale datasets, easily surpassing the most sophisticated state-of-the-art methods [38, 27, 2], by a large margin. Our XBM is conceptually simple, easy to implement - using several lines of codes, and is memory efficient - with a negligible 0.2 GB extra GPU memory.

The ongoing success of visual question answering methods has been somwehat surprising given that, at its most general, the problem requires understanding the entire variety of both visual and language stimuli. It is particularly remarkable that this success has been achieved on the basis of comparatively small datasets, given the scale of the problem. One explanation is that this has been accomplished partly by exploiting bias in the datasets rather than developing deeper multi-modal reasoning. This fundamentally limits the generalization of the method, and thus its practical applicability. We propose a method that addresses this problem by introducing counterfactuals in the training. In doing so we leverage structural causal models for counterfactual evaluation to formulate alternatives, for instance, questions that could be asked of the same image set. We show that simulating plausible alternative training data through this process results in better generalization.

We propose a cost volume-based neural network for depth inference from multi-view images. We demonstrate that building a cost volume pyramid in a coarse-to-fine manner instead of constructing a cost volume at a fixed resolution leads to a compact, lightweight network and allows us inferring high resolution depth maps to achieve better reconstruction results. To this end, we first build a cost volume based on uniform sampling of fronto-parallel planes across the entire depth range at the coarsest resolution of an image. Then, given current depth estimate, we construct new cost volumes iteratively on the pixelwise depth residual to perform depth map refinement. While sharing similar insight with Point-MVSNet as predicting and refining depth iteratively, we show that working on cost volume pyramid can lead to a more compact, yet efficient network structure compared with the Point-MVSNet on 3D points. We further provide detailed analyses of the relation between (residual) depth sampling and image resolution, which serves as a principle for building compact cost volume pyramid. Experimental results on benchmark datasets show that our model can perform 6x faster and has similar performance as state-of-the-art methods. Code is available at https://github.com/JiayuYANG/CVP-MVSNet

The single image super-resolution task is one of the most examined inverse problems in the past decade. In the recent years, Deep Neural Networks (DNNs) have shown superior performance over alternative methods when the acquisition process uses a fixed known downscaling kernel---typically a bicubic kernel. However, several recent works have shown that in practical scenarios, where the test data mismatch the training data (e.g. when the downscaling kernel is not the bicubic kernel or is not available at training), the leading DNN methods suffer from a huge performance drop. Inspired by the literature on generalized sampling, in this work we propose a method for improving the performance of DNNs that have been trained with a fixed kernel on observations acquired by other kernels. For a known kernel, we design a closed-form correction filter that modifies the low-resolution image to match one which is obtained by another kernel (e.g. bicubic), and thus improves the results of existing pre-trained DNNs. For an unknown kernel, we extend this idea and propose an algorithm for blind estimation of the required correction filter. We show that our approach outperforms other super-resolution methods, which are designed for general downscaling kernels.

Existing face hallucination methods based on convolutional neural networks (CNN) have achieved impressive performance on low-resolution (LR) faces in a normal illumination condition. However, their performance degrades dramatically when LR faces are captured in low or non-uniform illumination conditions. This paper proposes a Copy and Paste Generative Adversarial Network (CPGAN) to recover authentic high-resolution (HR) face images while compensating for low and non-uniform illumination. To this end, we develop two key components in our CPGAN: internal and external Copy and Paste nets (CPnets). Specifically, our internal CPnet exploits facial information residing in the input image to enhance facial details; while our external CPnet leverages an external HR face for illumination compensation. A new illumination compensation loss is thus developed to capture illumination from the external guided face image effectively. Furthermore, our method offsets illumination and upsamples facial details alternatively in a coarse-to-fine fashion, thus alleviating the correspondence ambiguity between LR inputs and external HR inputs. Extensive experiments demonstrate that our method manifests authentic HR face images in a uniform illumination condition and outperforms state-of-the-art methods qualitatively and quantitatively.