CVPR 2020 论文列表

We propose a novel quaternion product unit (QPU) to represent data on 3D rotation groups. The QPU leverages quaternion algebra and the law of 3D rotation group, representing 3D rotation data as quaternions and merging them via a weighted chain of Hamilton products. We prove that the representations derived by the proposed QPU can be disentangled into "rotation-invariant" features and "rotation-equivariant" features, respectively, which supports the rationality and the efficiency of the QPU in theory. We design quaternion neural networks based on our QPUs and make our models compatible with existing deep learning models. Experiments on both synthetic and real-world data show that the proposed QPU is beneficial for the learning tasks requiring rotation robustness.

Machine learning (ML), especially deep neural networks (DNNs) have been widely used in various applications, including several safety-critical ones (e.g. autonomous driving). As a result, recent research about adversarial examples has raised great concerns. Such adversarial attacks can be achieved by adding a small magnitude of perturbation to the input to mislead model prediction. While several whitebox attacks have demonstrated their effectiveness, which assume that the attackers have full access to the machine learning models; blackbox attacks are more realistic in practice. In this paper, we propose a Query-Efficient Boundary-based blackbox Attack (QEBA) based only on model's final prediction labels. We theoretically show why previous boundary-based attack with gradient estimation on the whole gradient space is not efficient in terms of query numbers, and provide optimality analysis for our dimension reduction-based gradient estimation. On the other hand, we conducted extensive experiments on ImageNet and CelebA datasets to evaluate QEBA. We show that compared with the state-of-the-art blackbox attacks, QEBA is able to use a smaller number of queries to achieve a lower magnitude of perturbation with 100% attack success rate. We also show case studies of attacks on real-world APIs including MEGVII Face++ and Microsoft Azure.

Context plays an important role in visual recognition. Recent studies have shown that visual recognition networks can be fooled by placing objects in inconsistent contexts (e.g., a cow in the ocean). To model the role of contextual information in visual recognition, we systematically investigated ten critical properties of where, when, and how context modulates recognition, including the amount of context, context and object resolution, geometrical structure of context, context congruence, and temporal dynamics of contextual modulation. The tasks involved recognizing a target object surrounded with context in a natural image. As an essential benchmark, we conducted a series of psychophysics experiments where we altered one aspect of context at a time, and quantified recognition accuracy. We propose a biologically-inspired context-aware object recognition model consisting of a two-stream architecture. The model processes visual information at the fovea and periphery in parallel, dynamically incorporates object and contextual information, and sequentially reasons about the class label for the target object. Across a wide range of behavioral tasks, the model approximates human level performance without retraining for each task, captures the dependence of context enhancement on image properties, and provides initial steps towards integrating scene and object information for visual recognition. All source code and data are publicly available: https://github.com/kreimanlab/Put-In-Context.

Portrait animation, which aims to animate a still portrait to life using poses extracted from target frames, is an important technique for many real-world entertainment applications. Although recent works have achieved highly realistic results on synthesizing or controlling human head images, the puppeteering of arbitrary portraits is still confronted by the following challenges: 1) identity/personality mismatch; 2) training data/domain limitations; and 3) low-efficiency in training/fine-tuning. In this paper, we devised a novel two-stage framework called PuppeteerGAN for solving these challenges. Specifically, we first learn identity-preserved semantic segmentation animation which executes pose retargeting between any portraits. As a general representation, the semantic segmentation results could be adapted to different datasets, environmental conditions or appearance domains. Furthermore, the synthesized semantic segmentation is filled with the appearance of the source portrait. To this end, an appearance transformation network is presented to produce fidelity output by jointly considering the wrapping of semantic features and conditional generation. After training, the two networks can directly perform end-to-end inference on unseen subjects without any retraining or fine-tuning. Extensive experiments on cross-identity/domain/resolution situations demonstrate the superiority of the proposed PuppetterGAN over existing portrait animation methods in both generation quality and inference speed.

Existing metric learning losses can be categorized into two classes: pair-based and proxy-based losses. The former class can leverage fine-grained semantic relations between data points, but slows convergence in general due to its high training complexity. In contrast, the latter class enables fast and reliable convergence, but cannot consider the rich data-to-data relations. This paper presents a new proxy-based loss that takes advantages of both pair- and proxy-based methods and overcomes their limitations. Thanks to the use of proxies, our loss boosts the speed of convergence and is robust against noisy labels and outliers. At the same time, it allows embedding vectors of data to interact with each other in its gradients to exploit data-to-data relations. Our method is evaluated on four public benchmarks, where a standard network trained with our loss achieves state-of-the-art performance and most quickly converges.

Deep learning technique has dramatically boosted the performance of face alignment algorithms. However, due to large variability and lack of samples, the alignment problem in unconstrained situations, e.g. large head poses, exaggerated expression, and uneven illumination, is still largely unsolved. In this paper, we explore the instincts and reasons behind our two proposals, i.e. Propagation Module and Focal Wing Loss, to tackle the problem. Concretely, we present a novel structure-infused face alignment algorithm based on heatmap regression via propagating landmark heatmaps to boundary heatmaps, which provide structure information for further attention map generation. Moreover, we propose a Focal Wing Loss for mining and emphasizing the difficult samples under in-the-wild condition. In addition, we adopt methods like CoordConv and Anti-aliased CNN from other fields that address the shift variance problem of CNN for face alignment. When implementing extensive experiments on different benchmarks, i.e. WFLW, 300W, and COFW, our method outperforms the state-of-the-arts by a significant margin. Our proposed approach achieves 4.05% mean error on WFLW, 2.93% mean error on 300W full-set, and 3.71% mean error on COFW.

Generating adversarial examples in a black-box setting retains a significant challenge with vast practical application prospects. In particular, existing black-box attacks suffer from the need for excessive queries, as it is non-trivial to find an appropriate direction to optimize in the high-dimensional space. In this paper, we propose Projection & Probability-driven Black-box Attack (PPBA) to tackle this problem by reducing the solution space and providing better optimization. For reducing the solution space, we first model the adversarial perturbation optimization problem as a process of recovering frequency-sparse perturbations with compressed sensing, under the setting that random noise in the low-frequency space is more likely to be adversarial. We then propose a simple method to construct a low-frequency constrained sensing matrix, which works as a plug-and-play projection matrix to reduce the dimensionality. Such a sensing matrix is shown to be flexible enough to be integrated into existing methods like NES and Bandits_ TD . For better optimization, we perform a random walk with a probability-driven strategy, which utilizes all queries over the whole progress to make full use of the sensing matrix for a less query budget. Extensive experiments show that our method requires at most 24% fewer queries with a higher attack success rate compared with state-of-the-art approaches. Finally, the attack method is evaluated on the real-world online service, i.e., Google Cloud Vision API, which further demonstrates our practical potentials.

Group activities usually involve spatio-temporal dynamics among many interactive individuals, while only a few participants at several key frames essentially define the activity. Therefore, effectively modeling the group-relevant and suppressing the irrelevant actions (and interactions) are vital for group activity recognition. In this paper, we propose a novel method based on deep reinforcement learning to progressively refine the low-level features and high-level relations of group activities. Firstly, we construct a semantic relation graph (SRG) to explicitly model the relations among persons. Then, two agents adopting policy according to two Markov decision processes are applied to progressively refine the SRG. Specifically, one feature-distilling (FD) agent in the discrete action space refines the low-level spatio-temporal features by distilling the most informative frames. Another relation-gating (RG) agent in continuous action space adjusts the high-level semantic graph to pay more attention to group-relevant relations. The SRG, FD agent, and RG agent are optimized alternately to mutually boost the performance of each other. Extensive experiments on two widely used benchmarks demonstrate the effectiveness and superiority of the proposed approach.

The mirror detection problem is important as mirrors can affect the performances of many vision tasks. It is a difficult problem as it requires an understanding of global scene semantics. Recently, a method was proposed to detect mirrors by learning multi-level contextual contrasts between inside and outside of mirrors, which helps locate mirror edges implicitly. We observe that the content of a mirror reflects the content of its surrounding, separated by the edge of the mirror. Hence, we propose a model in this paper to progressively learn the content similarity between the inside and outside of the mirror while explicitly detecting the mirror edges. Our work has two main contributions. First, we propose a new relational contextual contrasted local (RCCL) module to extract and compare the mirror features with its corresponding context features, and an edge detection and fusion (EDF) module to learn the features of mirror edges in complex scenes via explicit supervision. Second, we construct a challenging benchmark dataset of 6,461 mirror images. Unlike the existing MSD dataset, which has limited diversity, our dataset covers a variety of scenes and is much larger in scale. Experimental results show that our model outperforms relevant state-of-the-art methods.

Fine-grained visual categorization has long been considered as an important problem, however, its real application is still restricted, since precisely annotating a large fine-grained image dataset is a laborious task and requires expert-level human knowledge. A solution to this problem is applying domain adaptation approaches to fine-grained scenarios, where the key idea is to discover the commonality between existing fine-grained image datasets and massive unlabeled data in the wild. The main technical bottleneck lies in that the large inter-domain variation will deteriorate the subtle boundaries of small inter-class variation during domain alignment. This paper presents the Progressive Adversarial Networks (PAN) to align fine-grained categories across domains with a curriculum-based adversarial learning framework. In particular, throughout the learning process, domain adaptation is carried out through all multi-grained features, progressively exploiting the label hierarchy from coarse to fine. The progressive learning is applied upon both category classification and domain alignment, boosting both the discriminability and the transferability of the fine-grained features. Our method is evaluated on three benchmarks, two of which are proposed by us, and it outperforms the state-of-the-art domain adaptation methods.

Domain adaptive image retrieval includes single-domain retrieval and cross-domain retrieval. Most of the existing image retrieval methods only focus on single-domain retrieval, which assumes that the distributions of retrieval databases and queries are similar. However, in practical application, the discrepancies between retrieval databases often taken in ideal illumination/pose/background/camera conditions and queries usually obtained in uncontrolled conditions are very large. In this paper, considering the practical application, we focus on challenging cross-domain retrieval. To address the problem, we propose an effective method named Probability Weighted Compact Feature Learning (PWCF), which provides inter-domain correlation guidance to promote cross-domain retrieval accuracy and learns a series of compact binary codes to improve the retrieval speed. First, we derive our loss function through the Maximum A Posteriori Estimation (MAP): Bayesian Perspective (BP) induced focal-triplet loss, BP induced quantization loss and BP induced classification loss. Second, we propose a common manifold structure between domains to explore the potential correlation across domains. Considering the original feature representation is biased due to the inter-domain discrepancy, the manifold structure is difficult to be constructed. Therefore, we propose a new feature named Histogram Feature of Neighbors (HFON) from the sample statistics perspective. Extensive experiments on various benchmark databases validate that our method outperforms many state-of-the-art image retrieval methods for domain adaptive image retrieval. The source code is available at https://github.com/fuxianghuang1/PWCF .

We study a new research problem of probabilistic future frames prediction from a sequence of noisy inputs, which is useful because it is difficult to guarantee the quality of input frames in practical spatiotemporal prediction applications. It is also challenging because it involves two levels of uncertainty: the perceptual uncertainty from noisy observations and the dynamics uncertainty in forward modeling. In this paper, we propose to tackle this problem with an end-to-end trainable model named Bayesian Predictive Network (BP-Net). Unlike previous work in stochastic video prediction that assumes spatiotemporal coherence and therefore fails to deal with perceptual uncertainty, BP-Net models both levels of uncertainty in an integrated framework. Furthermore, unlike previous work that can only provide unsorted estimations of future frames, BP-Net leverages a differentiable sequential importance sampling (SIS) approach to make future predictions based on the inference of underlying physical states, thereby providing sorted prediction candidates in accordance with the SIS importance weights, i.e., the confidences. Our experiment results demonstrate that BP-Net remarkably outperforms existing approaches on predicting future frames from noisy data.

Unsupervised embedding learning aims at extracting low-dimensional visually meaningful representations from large-scale unlabeled images, which can then be directly used for similarity-based search. This task faces two major challenges: 1) mining positive supervision from highly similar fine-grained classes and 2) generating to unseen testing categories. To tackle these issues, this paper proposes a probabilistic structural latent representation (PSLR), which incorporates an adaptable softmax embedding to approximate the positive concentrated and negative instance separated properties in the graph latent space. It improves the discriminability by enlarging the positive/negative difference without introducing any additional computational cost while maintaining high learning efficiency. To address the limited supervision using data augmentation, a smooth variational reconstruction loss is introduced by modeling the intra-instance variance, which improves the robustness. Extensive experiments demonstrate the superiority of PSLR over state-of-the-art unsupervised methods on both seen and unseen categories with cosine similarity. Code is available at https://github.com/mangye16/PSLR

Visual tracking is fundamentally the problem of regressing the state of the target in each video frame. While significant progress has been achieved, trackers are still prone to failures and inaccuracies. It is therefore crucial to represent the uncertainty in the target estimation. Although current prominent paradigms rely on estimating a state-dependent confidence score, this value lacks a clear probabilistic interpretation, complicating its use. In this work, we therefore propose a probabilistic regression formulation and apply it to tracking. Our network predicts the conditional probability density of the target state given an input image. Crucially, our formulation is capable of modeling label noise stemming from inaccurate annotations and ambiguities in the task. The regression network is trained by minimizing the Kullback-Leibler divergence. When applied for tracking, our formulation not only allows a probabilistic representation of the output, but also substantially improves the performance. Our tracker sets a new state-of-the-art on six datasets, achieving 59.8% AUC on LaSOT and 75.8% Success on TrackingNet. The code and models are available at https://github.com/visionml/pytracking.

Encoder-decoder networks have found widespread use in various dense prediction tasks. However, the strong reduction of spatial resolution in the encoder leads to a loss of location information as well as boundary artifacts. To address this, image-adaptive post-processing methods have shown beneficial by leveraging the high-resolution input image(s) as guidance data. We extend such approaches by considering an important orthogonal source of information: the network's confidence in its own predictions. We introduce probabilistic pixel-adaptive convolutions (PPACs), which not only depend on image guidance data for filtering, but also respect the reliability of per-pixel predictions. As such, PPACs allow for image-adaptive smoothing and simultaneously propagating pixels of high confidence into less reliable regions, while respecting object boundaries. We demonstrate their utility in refinement networks for optical flow and semantic segmentation, where PPACs lead to a clear reduction in boundary artifacts. Moreover, our proposed refinement step is able to substantially improve the accuracy on various widely used benchmarks.

Contour shape alignment is a fundamental but challenging problem in computer vision, especially when the observations are partial, noisy, and largely misaligned. Recent ConvNet-based architectures that were proposed to align image structures tend to fail with contour representation of shapes, mostly due to the use of proximity-insensitive pixel-wise similarity measures as loss functions in their training processes. This work presents a novel ConvNet, "ProAlignNet," that accounts for large scale misalignments and complex transformations between the contour shapes. It infers the warp parameters in a multi-scale fashion with progressively increasing complex transformations over increasing scales. It learns --without supervision-- to align contours, agnostic to noise and missing parts, by training with a novel loss function which is derived an upperbound of a proximity-sensitive and local shape-dependent similarity metric that uses classical Morphological Chamfer Distance Transform. We evaluate the reliability of these proposals on a simulated MNIST noisy contours dataset via some basic sanity check experiments. Next, we demonstrate the effectiveness of the proposed models in two real-world applications of (i) aligning geo-parcel data to aerial image maps and (ii) refining coarsely annotated segmentation labels. In both applications, the proposed models consistently perform superior to state-of-the-art methods.

With increasing ethical and legal concerns on privacy for deep models in visual recognition, differential privacy has emerged as a mechanism to disguise membership of sensitive data in training datasets. Recent methods like Private Aggregation of Teacher Ensembles (PATE) leverage a large ensemble of teacher models trained on disjoint subsets of private data, to transfer knowledge to a student model with privacy guarantees. However, labeled vision data is often expensive and datasets, when split into many disjoint training sets, lead to significantly sub-optimal accuracy and thus hardly sustain good privacy bounds. We propose a practically data-efficient scheme based on private release of k-nearest neighbor (kNN) queries, which altogether avoids splitting the training dataset. Our approach allows the use of privacy-amplification by subsampling and iterative refinement of the kNN feature embedding. We rigorously analyze the theoretical properties of our method and demonstrate strong experimental performance on practical computer vision datasets for face attribute recognition and person reidentification. In particular, we achieve comparable or better accuracy than PATE while reducing more than 90% of the privacy loss, thereby providing the "most practical method to-date" for private deep learning in computer vision.

Contemporary deep learning based semantic inpainting can be approached from two directions. First, and the more explored, approach is to train an offline deep regression network over the masked pixels with an additional refinement by adversarial training. This approach requires a single feed-forward pass for inpainting at inference. Another promising, yet unexplored approach is to first train a generative model to map a latent prior distribution to natural image manifold and during inference time search for the best-matching prior to reconstruct the signal. The primary aversion towards the latter genre is due to its inference time iterative optimization and difficulty to scale to higher resolution. In this paper, going against the general trend, we focus on the second paradigm of inpainting and address both of its mentioned problems. Most importantly, we learn a data driven parametric network to directly predict a matching prior for a given masked image. This converts an iterative paradigm to a single feed forward inference pipeline with around 800X speedup. We also regularize our network with structural prior (computed from the masked image itself) which helps in better preservation of pose and size of the object to be inpainted. Moreover, to extend our model for sequence reconstruction, we propose a recurrent net based grouped latent prior learning. Finally, we leverage recent advancements in high resolution GAN training to scale our inpainting network to 256X256. Experiments (spanning across resolutions from 64X64 to 256X256) conducted on SVHN, Standford Cars, CelebA, CelebA-HQ and ImageNet image datasets, and FaceForensics video datasets reveal that we consistently improve upon contemporary benchmarks from both schools of approaches.

It is a common paradigm in object detection frameworks to treat all samples equally and target at maximizing the performance on average. In this work, we revisit this paradigm through a careful study on how different samples contribute to the overall performance measured in terms of mAP. Our study suggests that the samples in each mini-batch are neither independent nor equally important, and therefore a better classifier on average does not necessarily result in higher mAP. Motivated by this study, we propose the notion of Prime Samples, those that play a key role in driving the detection performance. We further develop a simple yet effective sampling and learning strategy called PrIme Sample Attention (PISA) that directs the focus of the training process towards such samples. Our experiments demonstrate that it is often more effective to focus on prime samples than hard samples when training a detector. Particularly, on the MSCOCO dataset, PISA outperforms the random sampling baseline and hard mining schemes, e.g. OHEM and Focal Loss, consistently by around 2% on both single-stage and two-stage detectors, even with a strong backbone ResNeXt-101. Code is available at: https://github.com/open-mmlab/mmdetection.