CVPR 2020 论文列表

The widespread deployment of deep models necessitates the assessment of model vulnerability in practice, especially for safety- and security-sensitive domains such as autonomous driving and medical diagnosis. Transfer-based attacks against image classifiers thus elicit mounting interest, where attackers are required to craft adversarial images based on local proxy models without the feedback information from remote target ones. However, under such a challenging but practical setup, the synthesized adversarial samples often achieve limited success due to overfitting to the local model employed. In this work, we propose a novel mechanism to alleviate the overfitting issue. It computes model attention over extracted features to regularize the search of adversarial examples, which prioritizes the corruption of critical features that are likely to be adopted by diverse architectures. Consequently, it can promote the transferability of resultant adversarial instances. Extensive experiments on ImageNet classifiers confirm the effectiveness of our strategy and its superiority to state-of-the-art benchmarks in both white-box and black-box settings.

Few-shot learning (FSL) has attracted increasing attention in recent years but remains challenging, due to the intrinsic difficulty in learning to generalize from a few examples. This paper proposes an adaptive margin principle to improve the generalization ability of metric-based meta-learning approaches for few-shot learning problems. Specifically, we first develop a class-relevant additive margin loss, where semantic similarity between each pair of classes is considered to separate samples in the feature embedding space from similar classes. Further, we incorporate the semantic context among all classes in a sampled training task and develop a task-relevant additive margin loss to better distinguish samples from different classes. Our adaptive margin method can be easily extended to a more realistic generalized FSL setting. Extensive experiments demonstrate that the proposed method can boost the performance of current metric-based meta-learning approaches, under both the standard FSL and generalized FSL settings.

Neural Architecture Search (NAS), aiming at automatically designing network architectures by machines, is expected to bring about a new revolution in machine learning. Despite these high expectation, the effectiveness and efficiency of existing NAS solutions are unclear, with some recent works going so far as to suggest that many existing NAS solutions are no better than random architecture selection. The ineffectiveness of NAS solutions may be attributed to inaccurate architecture evaluation. Specifically, to speed up NAS, recent works have proposed under-training different candidate architectures in a large search space concurrently by using shared network parameters; however, this has resulted in incorrect architecture ratings and furthered the ineffectiveness of NAS. In this work, we propose to modularize the large search space of NAS into blocks to ensure that the potential candidate architectures are fully trained; this reduces the representation shift caused by the shared parameters and leads to the correct rating of the candidates. Thanks to the block-wise search, we can also evaluate all of the candidate architectures within each block. Moreover, we find that the knowledge of a network model lies not only in the network parameters but also in the network architecture. Therefore, we propose to distill the neural architecture (DNA) knowledge from a teacher model to supervise our block-wise architecture search, which significantly improves the effectiveness of NAS. Remarkably, the performance of our searched architectures has exceeded the teacher model, demonstrating the practicability of our method. Finally, our method achieves a state-of-the-art 78.4% top-1 accuracy on ImageNet in a mobile setting. All of our searched models along with the evaluation code are available at https://github.com/changlin31/DNA.

Blind image quality assessment (BIQA) for authentically distorted images has always been a challenging problem, since images captured in the wild include varies contents and diverse types of distortions. The vast majority of prior BIQA methods focus on how to predict synthetic image quality, but fail when applied to real-world distorted images. To deal with the challenge, we propose a self-adaptive hyper network architecture to blind assess image quality in the wild. We separate the IQA procedure into three stages including content understanding, perception rule learning and quality predicting. After extracting image semantics, perception rule is established adaptively by a hyper network, and then adopted by a quality prediction network. In our model, image quality can be estimated in a self-adaptive manner, thus generalizes well on diverse images captured in the wild. Experimental results verify that our approach not only outperforms the state-of-the-art methods on challenging authentic image databases but also achieves competing performances on synthetic image databases, though it is not explicitly designed for the synthetic task.

While deep learning has recently achieved great success on multi-view stereo (MVS), limited training data makes the trained model hard to be generalized to unseen scenarios. Compared with other computer vision tasks, it is rather difficult to collect a large-scale MVS dataset as it requires expensive active scanners and labor-intensive process to obtain ground truth 3D structures. In this paper, we introduce BlendedMVS, a novel large-scale dataset, to provide sufficient training ground truth for learning-based MVS. To create the dataset, we apply a 3D reconstruction pipeline to recover high-quality textured meshes from images of well-selected scenes. Then, we render these mesh models to color images and depth maps. To introduce the ambient lighting information during training, the rendered color images are further blended with the input images to generate the training input. Our dataset contains over 17k high-resolution images covering a variety of scenes, including cities, architectures, sculptures and small objects. Extensive experiments demonstrate that BlendedMVS endows the trained model with significantly better generalization ability compared with other MVS datasets. The dataset and pretrained models are available at https://github.com/YoYo000/BlendedMVS.

Binary Neural Networks (BNNs), known to be one among the effectively compact network architectures, have achieved great outcomes in the visual tasks. Designing efficient binary architectures is not trivial due to the binary nature of the network. In this paper, we propose a use of evolutionary search to facilitate the construction and training scheme when binarizing MobileNet, a compact network with separable depth-wise convolution. Being inspired by one-shot architecture search frameworks, we manipulate the idea of group convolution to design efficient 1-Bit Convolutional Neural Networks (CNNs), assuming an approximately optimal trade-off between computational cost and model accuracy. Our objective is to come up with a tiny yet efficient binary neural architecture by exploring the best candidates of the group convolution while optimizing the model performance in terms of complexity and latency. The approach is threefold. First, we modify and train strong baseline binary networks with a wide range of random group combinations at each convolutional layer. This set-up gives the binary neural networks a capability of preserving essential information through layers. Second, to find a good set of hyper-parameters for group convolutions we make use of the evolutionary search which leverages the exploration of efficient 1-bit models. Lastly, these binary models are trained from scratch in a usual manner to achieve the final binary model. Various experiments on ImageNet are conducted to show that following our construction guideline, the final model achieves 60.09% Top-1 accuracy and outperforms the state-of-the-art CI-BCNN with the same computational cost.

Recent researches on panoptic segmentation resort to a single end-to-end network to combine the tasks of instance segmentation and semantic segmentation. However, prior models only unified the two related tasks at the architectural level via a multi-branch scheme or revealed the underlying correlation between them by unidirectional feature fusion, which disregards the explicit semantic and co-occurrence relations among objects and background. Inspired by the fact that context information is critical to recognize and localize the objects, and inclusive object details are significant to parse the background scene, we thus investigate on explicitly modeling the correlations between object and background to achieve a holistic understanding of an image in the panoptic segmentation task. We introduce a Bidirectional Graph Reasoning Network (BGRNet), which incorporates graph structure into the conventional panoptic segmentation network to mine the intra-modular and inter-modular relations within and between foreground things and background stuff classes. In particular, BGRNet first constructs image-specific graphs in both instance and semantic segmentation branches that enable flexible reasoning at the proposal level and class level, respectively. To establish the correlations between separate branches and fully leverage the complementary relations between things and stuff, we propose a Bidirectional Graph Connection Module to diffuse information across branches in a learnable fashion. Experimental results demonstrate the superiority of our BGRNet that achieves the new state-of-the-art performance on challenging COCO and ADE20K panoptic segmentation benchmarks.

Heavy haze results in severe image degradation and thus hampers the performance of visual perception, object detection, etc. On the assumption that dehazed binocular images are superior to the hazy ones for stereo vision tasks such as 3D object detection and according to the fact that image haze is a function of depth, this paper proposes a Binocular image dehazing Network (BidNet) aiming at dehazing both the left and right images of binocular images within the deep learning framework. Existing binocular dehazing methods rely on simultaneously dehazing and estimating disparity, whereas BidNet does not need to explicitly perform time-consuming and well-known challenging disparity estimation. Note that a small error in disparity gives rise to a large variation in depth and in estimation of haze-free image. The relationship and correlation between binocular images are explored and encoded by the proposed Stereo Transformation Module (STM). Jointly dehazing binocular image pairs is mutually beneficial, which is better than only dehazing left images. We extend the Foggy Cityscapes dataset to a Stereo Foggy Cityscapes dataset with binocular foggy image pairs. Experimental results demonstrate that BidNet significantly outperforms state-of-the-art dehazing methods in both subjective and objective assessments.

Depth estimation from a monocular 360 image is an emerging problem that gains popularity due to the availability of consumer-level 360 cameras and the complete surrounding sensing capability. While the standard of 360 imaging is under rapid development, we propose to predict the depth map of a monocular 360 image by mimicking both peripheral and foveal vision of the human eye. To this end, we adopt a two-branch neural network leveraging two common projections: equirectangular and cubemap projections. In particular, equirectangular projection incorporates a complete field-of-view but introduces distortion, whereas cubemap projection avoids distortion but introduces discontinuity at the boundary of the cube. Thus we propose a bi-projection fusion scheme along with learnable masks to balance the feature map from the two projections. Moreover, for the cubemap projection, we propose a spherical padding procedure which mitigates discontinuity at the boundary of each face. We apply our method to four panorama datasets and show favorable results against the existing state-of-the-art methods.

In this paper, we propose a binarized neural network learning method called BiDet for efficient object detection. Conventional network binarization methods directly quantize the weights and activations in one-stage or two-stage detectors with constrained representational capacity, so that the information redundancy in the networks causes numerous false positives and degrades the performance significantly. On the contrary, our BiDet fully utilizes the representational capacity of the binary neural networks for object detection by redundancy removal, through which the detection precision is enhanced with alleviated false positives. Specifically, we generalize the information bottleneck (IB) principle to object detection, where the amount of information in the high-level feature maps is constrained and the mutual information between the feature maps and object detection is maximized. Meanwhile, we learn sparse object priors so that the posteriors are concentrated on informative detection prediction with false positive elimination. Extensive experiments on the PASCAL VOC and COCO datasets show that our method outperforms the state-of-the-art binary neural networks by a sizable margin.

Stereo-based depth estimation is a cornerstone of computer vision, with state-of-the-art methods delivering accurate results in real time. For several applications such as autonomous navigation, however, it may be useful to trade accuracy for lower latency. We present Bi3D, a method that estimates depth via a series of binary classifications. Rather than testing if objects are at a particular depth D, as existing stereo methods do, it classifies them as being closer or farther than D. This property offers a powerful mechanism to balance accuracy and latency. Given a strict time budget, Bi3D can detect objects closer than a given distance in as little as a few milliseconds, or estimate depth with arbitrarily coarse quantization, with complexity linear with the number of quantization levels. Bi3D can also use the allotted quantization levels to get continuous depth, but in a specific depth range. For standard stereo (i.e., continuous depth on the whole range), our method is close to or on par with state-of-the-art, finely tuned stereo methods.

Most existing methods do not explicitly formulate the mutual guidance between vision and language. In this work, we propose a bi-directional relationship inferring network (BRINet) to model the dependencies of cross-modal information. In detail, the vision-guided linguistic attention is used to learn the adaptive linguistic context corresponding to each visual region. Combining with the language-guided visual attention, a bi-directional cross-modal attention module (BCAM) is built to learn the relationship between multi-modal features. Thus, the ultimate semantic context of the target object and referring expression can be represented accurately and consistently. Moreover, a gated bi-directional fusion module (GBFM) is designed to integrate the multi-level features where a gate function is used to guide the bi-directional flow of multi-level information. Extensive experiments on four benchmark datasets demonstrate that the proposed method outperforms other state-of-the-art methods under different evaluation metrics.

Existing works have designed end-to-end frameworks based on Faster-RCNN for person search. Due to the large receptive fields in deep networks, the feature maps of each proposal, cropped from the stem feature maps, involve redundant context information outside the bounding boxes. However, person search is a fine-grained task which needs accurate appearance information. Such context information can make the model fail to focus on persons, so the learned representations lack the capacity to discriminate various identities. To address this issue, we propose a Siamese network which owns an additional instance-aware branch, named Bi-directional Interaction Network (BINet). During the training phase, in addition to scene images, BINet also takes as inputs person patches which help the model discriminate identities based on human appearance. Moreover, two interaction losses are designed to achieve bi-directional interaction between branches at two levels. The interaction can help the model learn more discriminative features for persons in the scene. At the inference stage, only the major branch is applied, so BINet introduces no additional computation. Extensive experiments on two widely used person search benchmarks, CUHK-SYSU and PRW, have shown that our BINet achieves state-of-the-art results among end-to-end methods without loss of efficiency.

Video visual relation detection (VidVRD) aims to describe all interacting objects in a video. Different from relationships in static images, videos contain an addition temporal channel. A majority of existing works divide a video into short segments, predict relationships in each segment, and merge them. Such methods cannot capture relations involving long motions. Predicting the same relationship across neighboring video segments is also inefficient. To address these issues, this work proposes a novel sliding-window scheme to simultaneously predict short-term and long-term relationships. We run windows with different kernel sizes on object tracklets to generate sub-tracklet proposals with different duration, while the computational load is similar to that in segment-based methods. To fully utilize spatial and temporal information in videos, we construct one spatial and one temporal graph and employ Graph Convloutional Network to generate contextual embedding for tracklet proposal compatibility evaluation. We only predict relationships on highly-compatible proposal pairs. Our method achieves state-of-the-art performance on both ImageNet-VidVRD and VidOR dataset across multiple tasks. Especially for ImageNet-VidVRD, we obtain an average of 3% (R@50 from 8.07% to 11.21%) improvement under all evaluation metrics.

Sequence-level learning objective has been widely used in captioning tasks to achieve the state-of-the-art performance for many models. In this objective, the model is trained by the reward on the quality of its generated captions (sequence-level). In this work, we show the limitation of the current sequence-level learning objective for captioning tasks from both theory and empirical result. In theory, we show that the current objective is equivalent to only optimizing the precision side of the caption set generated by the model and therefore overlooks the recall side. Empirical result shows that the model trained by this objective tends to get lower score on the recall side. We propose to add a sequence-level exploration term to the current objective to boost recall. It guides the model to explore more plausible captions in the training. In this way, the proposed objective takes both the precision and recall sides of generated captions into account. Experiments show the effectiveness of the proposed method on both video and image captioning datasets.

When designing a semantic segmentation module for a practical application, such as autonomous driving, it is crucial to understand the robustness of the module with respect to a wide range of image corruptions. While there are recent robustness studies for full-image classification, we are the first to present an exhaustive study for semantic segmentation, based on the state-of-the-art model DeepLabv3+. To increase the realism of our study, we utilize almost 400,000 images generated from Cityscapes, PASCAL VOC 2012, and ADE20K. Based on the benchmark study, we gain several new insights. Firstly, contrary to full-image classification, model robustness increases with model performance, in most cases. Secondly, some architecture properties affect robustness significantly, such as a Dense Prediction Cell, which was designed to maximize performance on clean data only.

It has been proposed by many researchers that combining deep neural networks with graphical models can create more efficient and better regularized composite models. The main difficulties in implementing this in practice are associated with a discrepancy in suitable learning objectives as well as with the necessity of approximations for the inference. In this work we take one of the simplest inference methods, a truncated max-product Belief Propagation, and add what is necessary to make it a proper component of a deep learning model: connect it to learning formulations with losses on marginals and compute the backprop operation. This BP-Layer can be used as the final or an intermediate block in convolutional neural networks (CNNs), allowing us to design a hierarchical model composing BP inference and CNNs at different scale levels. The model is applicable to a range of dense prediction problems, is well-trainable and provides parameter-efficient and robust solutions in stereo, flow and semantic segmentation.

Bursts of images exhibit significant self-similarity across both time and space. This motivates a representation of the kernels as linear combinations of a small set of basis elements. To this end, we introduce a novel basis prediction network that, given an input burst, predicts a set of global basis kernels --- shared within the image --- and the corresponding mixing coefficients --- which are specific to individual pixels. Compared to state-of-the-art techniques that output a large tensor of per-pixel spatiotemporal kernels, our formulation substantially reduces the dimensionality of the network output. This allows us to effectively exploit comparatively larger denoising kernels, achieving both significant quality improvements (over 1dB PSNR) and faster run-times over state-of-the-art methods.

Image interpolation, or image morphing, refers to a visual transition between two (or more) input images. For such a transition to look visually appealing, its desirable properties are (i) to be smooth; (ii) to apply the minimal required change in the image; and (iii) to seem "real", avoiding unnatural artifacts in each image in the transition. To obtain a smooth and straightforward transition, one may adopt the well-known Wasserstein Barycenter Problem (WBP). While this approach guarantees minimal changes under the Wasserstein metric, the resulting images might seem unnatural. In this work, we propose a novel approach for image morphing that possesses all three desired properties. To this end, we define a constrained variant of the WBP that enforces the intermediate images to satisfy an image prior. We describe an algorithm that solves this problem and demonstrate it using the sparse prior and generative adversarial networks.