CVPR 2021 论文列表

This paper addresses the problem of 3D human body shape and pose estimation from RGB images. Recent progress in this field has focused on single images, video or multi-view images as inputs. In contrast, we propose a new task: shape and pose estimation from a group of multiple images of a human subject, without constraints on subject pose, camera viewpoint or background conditions between images in the group. Our solution to this task predicts distributions over SMPL body shape and pose parameters conditioned on the input images in the group. We probabilistically combine predicted body shape distributions from each image to obtain a final multi-image shape prediction. We show that the additional body shape information present in multi-image input groups improves 3D human shape estimation metrics compared to single-image inputs on the SSP-3D dataset and a private dataset of tape-measured humans. In addition, predicting distributions over 3D bodies allows us to quantify pose prediction uncertainty, which is useful when faced with challenging input images with significant occlusion. Our method demonstrates meaningful pose uncertainty on the 3DPW dataset and is competitive with the state-of-the-art in terms of pose estimation metrics.

To train robust deep neural networks (DNNs), we systematically study several target modification approaches, which include output regularisation, self and non-self label correction (LC). Two key issues are discovered: (1) Self LC is the most appealing as it exploits its own knowledge and requires no extra models. However, how to automatically decide the trust degree of a learner as training goes is not well answered in the literature? (2) Some methods penalise while the others reward low-entropy predictions, prompting us to ask which one is better? To resolve the first issue, taking two well-accepted propositions-deep neural networks learn meaningful patterns before fitting noise (Arpit et al., 2017) and minimum entropy regularisation principle (Grandvalet & Bengio, 2006)-we propose a novel end-to-end method named ProSelfLC, which is designed according to learning time and entropy. Specifically, given a data point, we progressively increase trust in its predicted label distribution versus its annotated one if a model has been trained for enough time and the prediction is of low entropy (high confidence). For the second issue, according to ProSelfLC, we empirically prove that it is better to redefine a meaningful low-entropy status and optimise the learner toward it. This serves as a defence of entropy minimisation. We demonstrate the effectiveness of ProSelfLC through extensive experiments in both clean and noisy settings. The source code is available at https://github.com/XinshaoAmosWang/ProSelfLC-CVPR2021.

Many computer vision systems require users to upload image features to the cloud for processing and storage. These features can be exploited to recover sensitive information about the scene or subjects, e.g., by reconstructing the appearance of the original image. To address this privacy concern, we propose a new privacy-preserving feature representation. The core idea of our work is to drop constraints from each feature descriptor by embedding it within an affine subspace containing the original feature as well as adversarial feature samples. Feature matching on the privacy-preserving representation is enabled based on the notion of subspace-to-subspace distance. We experimentally demonstrate the effectiveness of our method and its high practical relevance for the applications of visual localization and mapping as well as face authentication. Compared to the original features, our approach makes it significantly more difficult for an adversary to recover private information.

Collaborative learning has gained great popularity due to its benefit of data privacy protection: participants can jointly train a Deep Learning model without sharing their training sets. However, recent works discovered that an adversary can fully recover the sensitive training samples from the shared gradients. Such reconstruction attacks pose severe threats to collaborative learning. Hence, effective mitigation solutions are urgently desired. In this paper, we propose to leverage data augmentation to defeat reconstruction attacks: by preprocessing sensitive images with carefully-selected transformation policies, it becomes infeasible for the adversary to extract any useful information from the corresponding gradients. We design a novel search method to automatically discover qualified policies. We adopt two new metrics to quantify the impacts of transformations on data privacy and model usability, which can significantly accelerate the search speed. Comprehensive evaluations demonstrate that the policies discovered by our method can defeat existing reconstruction attacks in collaborative learning, with high efficiency and negligible impact on the model performance.

Recent works on localization and mapping from privacy preserving line features have made significant progress towards addressing the privacy concerns arising from cloud-based solutions in mixed reality and robotics. The requirement for calibrated cameras is a fundamental limitation for these approaches, which prevents their application in many crowd-sourced mapping scenarios. In this paper, we propose a solution to the uncalibrated privacy preserving localization and mapping problem. Our approach simultaneously recovers the intrinsic and extrinsic calibration of a camera from line-features only. This enables uncalibrated devices to both localize themselves within an existing map as well as contribute to the map, while preserving the privacy of the image contents. Furthermore, we also derive a solution to bootstrapping maps from scratch using only uncalibrated devices. Our approach provides comparable performance to the calibrated scenario and the privacy compromising alternatives based on traditional point features.

One-shot neural architecture search (NAS) methods significantly reduce the search cost by considering the whole search space as one network, which only needs to be trained once. However, current methods select each operation independently without considering previous layers. Besides, the historical information obtained with huge computation costs is usually used only once and then discarded. In this paper, we introduce a sampling strategy based on Monte Carlo tree search (MCTS) with the search space modeled as a Monte Carlo tree (MCT), which captures the dependency among layers. Furthermore, intermediate results are stored in the MCT for future decisions and a better exploration-exploitation balance. Concretely, MCT is updated using the training loss as a reward to the architecture performance; for accurately evaluating the numerous nodes, we propose node communication and hierarchical node selection methods in the training and search stages, respectively, making better uses of the operation rewards and hierarchical information. Moreover, for a fair comparison of different NAS methods, we construct an open-source NAS benchmark of a macro search space evaluated on CIFAR-10, namely NAS-Bench-Macro. Extensive experiments on NAS-Bench-Macro and ImageNet demonstrate that our method significantly improves search efficiency and performance. For example, by only searching 20 architectures, our obtained architecture achieves 78.0% top-1 accuracy with 442M FLOPs on ImageNet. Code (Benchmark) is available at: https://github.com/xiusu/NAS-Bench-Macro.

We study a very challenging task, human image completion, which tries to recover the human body part with a reasonable human shape from the corrupted region. Since each human body part is unique, it is infeasible to restore the missing part by borrowing textures from other visible regions. Thus, we propose two types of learned priors to compensate for the damaged region. One is a structure prior, it uses a human parsing map to represent the human body structure. The other is a structure-texture correlation prior. It learns a structure and a texture memory bank, which encodes the common body structures and texture patterns, respectively. With the aid of these memory banks, the model could utilize the visible pattern to query and fetch a similar structure and texture pattern to introduce additional reasonable structures and textures for the corrupted region. Besides, since multiple potential human shapes are underlying the corrupted region, we propose multi-scale structure discriminators to further restore a plausible topological structure. Experiments on various large-scale benchmarks demonstrate the effectiveness of our proposed method.

Scene text recognition is a challenging task due to diverse variations of text instances in natural scene images. Conventional methods based on CNN-RNN-CTC or encoder-decoder with attention mechanism may not fully investigate stable and efficient feature representations for multi-oriented scene texts. In this paper, we propose a primitive representation learning method that aims to exploit intrinsic representations of scene text images. We model elements in feature maps as the nodes of an undirected graph. A pooling aggregator and a weighted aggregator are proposed to learn primitive representations, which are transformed into high-level visual text representations by graph convolutional networks. A Primitive REpresentation learning Network (PREN) is constructed to use the visual text representations for parallel decoding. Furthermore, by integrating visual text representations into an encoder-decoder model with the 2D attention mechanism, we propose a framework called PREN2D to alleviate the misalignment problem in attention-based methods. Experimental results on both English and Chinese scene text recognition tasks demonstrate that PREN keeps a balance between accuracy and efficiency, while PREN2D achieves state-of-the-art performance.

Attention has been an important mechanism for both humans and computer vision systems. While state-of-the-art models to predict attention focus on estimating a static probabilistic saliency map with free-viewing behavior, real-life scenarios are filled with tasks of varying types and complexities, and visual exploration is a temporal process that contributes to task performance. To bridge the gap, we conduct a first study to understand and predict the temporal sequences of eye fixations (a.k.a. scanpaths) during performing general tasks, and examine how scanpaths affect task performance. We present a new deep reinforcement learning method to predict scanpaths leading to different performances in visual question answering. Conditioned on a task guidance map, the proposed model learns question-specific attention patterns to generate scanpaths. It addresses the exposure bias in scanpath prediction with self-critical sequence training and designs a Consistency-Divergence loss to generate distinguishable scanpaths between correct and incorrect answers. The proposed model not only accurately predicts the spatio-temporal patterns of human behavior in visual question answering, such as fixation position, duration, and order, but also generalizes to free-viewing and visual search tasks, achieving human-level performance in all tasks and significantly outperforming the state of the art.

We introduce PREDATOR, a model for pairwise pointcloud registration with deep attention to the overlap region. Different from previous work, our model is specifically designed to handle (also) point-cloud pairs with low overlap. Its key novelty is an overlap-attention block for early information exchange between the latent encodings of the two point clouds. In this way the subsequent decoding of the latent representations into per-point features is conditioned on the respective other point cloud, and thus can predict which points are not only salient, but also lie in the overlap region between the two point clouds. The ability to focus on points that are relevant for matching greatly improves performance: PREDATOR raises the rate of successful registrations by more than 20% in the low-overlap scenario, and also sets a new state of the art for the 3DMatch benchmark with 89% registration recall.

As the computing power of modern hardware is increasing strongly, pre-trained deep learning models (e.g., BERT, GPT-3) learned on large-scale datasets have shown their effectiveness over conventional methods. The big progress is mainly contributed to the representation ability of transformer and its variant architectures. In this paper, we study the low-level computer vision task (e.g., denoising, super-resolution and deraining) and develop a new pre-trained model, namely, image processing transformer (IPT). To maximally excavate the capability of transformer, we present to utilize the well-known ImageNet benchmark for generating a large amount of corrupted image pairs. The IPT model is trained on these images with multi-heads and multi-tails. In addition, the constructive learning is introduced for well adapting to different image processing tasks. The pre-trained model can therefore efficiently employed on desired task after fine-tuning. With only one pre-trained model, IPT outperforms the current state-of-the-art methods on various low-level benchmarks. Code is available at https://gitee.com/mindspore/mindspore/tree/master/model_zoo/research/cv/IPT

Wide-angle portraits often enjoy expanded views. However, they contain perspective distortions, especially noticeable when capturing group portrait photos, where the background is skewed and faces are stretched. This paper introduces the first deep learning based approach to remove such artifacts from freely-shot photos. Specifically, given a wide-angle portrait as input, we build a cascaded network consisting of a LineNet, a ShapeNet, and a transition module (TM), which corrects perspective distortions on the background, adapts to the stereographic projection on facial regions, and achieves smooth transitions between these two projections, accordingly. To train our network, we build the first perspective portrait dataset with a large diversity in identities, scenes and camera modules. For the quantitative evaluation, we introduce two novel metrics, line consistency and face congruence. Compared to the previous state-of-the-art approach, our method does not require camera distortion parameters. We demonstrate that our approach significantly outperforms the previous state-of-the-art approach both qualitatively and quantitatively.

A number of super-resolution (SR) algorithms from interpolation to deep neural networks (DNN) have emerged to restore or create missing details of the input low-resolution image. As mobile devices and display hardware develops, the demand for practical SR technology has increased. Current state-of-the-art SR methods are based on DNNs for better quality. However, they are feasible when executed by using a parallel computing module (e.g. GPUs), and have been difficult to apply to general uses such as end-user software, smartphones, and televisions. To this end, we propose an efficient and practical approach for the SR by adopting look-up table (LUT). We train a deep SR network with a small receptive field and transfer the output values of the learned deep model to the LUT. At test time, we retrieve the precomputed HR output values from the LUT for query LR input pixels. The proposed method can be performed very quickly because it does not require a large number of floating point operations. Experimental results show the efficiency and the effectiveness of our method. Especially, our method runs faster while showing better quality compared to bicubic interpolation.

Sufficient real information in generator is a critical point for the generation ability of GAN. However, GAN and its variants suffer from lack of this point, resulting in brittle training processes. In this paper, we propose a novel variant of GAN, Posterior Promoted GAN (P2GAN), which promotes generator with the real information in the posterior distribution produced by discriminator. In our framework, different from other variants of GAN, the discriminator maps images to a multivariate Gaussian distribution and extracts real information. The generator employs the real information by AdaIN and a latent code regularizer. Besides, reparameterization trick and pretraining is applied to guarantee a stable training process in practice. The convergence of P2GAN is theoretically proved. Experimental results on typical high-dimensional multi-modal datasets demonstrate that P2GAN has achieved comparable results with the state-of-the-art variants of GAN on unsupervised image synthesis.

We address the problem of uncertainty calibration. While standard deep neural networks typically yield uncalibrated predictions, calibrated confidence scores that are representative of the true likelihood of a prediction can be achieved using post-hoc calibration methods. However, to date, the focus of these approaches has been on in-domain calibration. Our contribution is two-fold. First, we show that existing post-hoc calibration methods yield highly over-confident predictions under domain shift. Second, we introduce a simple strategy where perturbations are applied to samples in the validation set before performing the post-hoc calibration step. In extensive experiments, we demonstrate that this perturbation step results in substantially better calibration under domain shift on a wide range of architectures and modelling tasks.

Deep learning based object detection approaches have achieved great progress with the benefit from large amount of labeled images. However, image annotation remains a laborious, time-consuming and error-prone process. To further improve the performance of detectors, we seek to exploit all available labeled data and excavate useful samples from massive unlabeled images in the wild, which is rarely discussed before. In this paper, we present a positive-unlabeled learning based scheme to expand training data by purifying valuable images from massive unlabeled ones, where the original training data are viewed as positive data and the unlabeled images in the wild are unlabeled data. To effectively utilized these purified data, we propose a self-distillation algorithm based on hint learning and ground truth bounded knowledge distillation. Experimental results verify that the proposed positive-unlabeled data purification can strengthen the original detector by mining the massive unlabeled data. In particular, our method boosts the mAP of FPN by +2.0% on COCO benchmark.

Reducing inconsistencies in the behavior of different versions of an AI system can be as important in practice as reducing its overall error. In image classification, sample-wise inconsistencies appear as "negative flips": A new model incorrectly predicts the output for a test sample that was correctly classified by the old (reference) model. Positive-congruent (PC) training aims at reducing error rate while at the same reducing negative flips, thus maximizing congruency with the reference model only on positive predictions, unlike model distillation. We propose a simple approach for PC training, Focal Distillation, which enforces congruence with the reference model by giving more weights to samples that were correctly classified. We also found that, if the reference model itself can be chosen as an ensemble of multiple deep neural networks, negative flips can be further reduced without affecting the new model's accuracy.

Visual and audio signals often coexist in natural environments, forming audio-visual events (AVEs). Given a video, we aim to localize video segments containing an AVE and identify its category. In order to learn discriminative features for a classifier, it is pivotal to identify the helpful (or positive) audio-visual segment pairs while filtering out the irrelevant ones, regardless whether they are synchronized or not. To this end, we propose a new positive sample propagation (PSP) module to discover and exploit the closely related audio-visual pairs by evaluating the relationship within every possible pair. It can be done by constructing an all-pair similarity map between each audio and visual segment, and only aggregating the features from the pairs with high similarity scores. To encourage the network to extract high correlated features for positive samples, a new audio-visual pair similarity loss is proposed. We also propose a new weighting branch to better exploit the temporal correlations in weakly supervised setting. We perform extensive experiments on the public AVE dataset and achieve new state-of-the-art accuracy in both fully and weakly supervised settings, thus verifying the effectiveness of our method.

SinGAN shows impressive capability in learning internal patch distribution despite its limited effective receptive field. We are interested in knowing how such a translation-invariant convolutional generator could capture the global structure with just a spatially i.i.d. input. In this work, taking SinGAN and StyleGAN2 as examples, we show that such capability, to a large extent, is brought by the implicit positional encoding when using zero padding in the generators. Such positional encoding is indispensable for generating images with high fidelity. The same phenomenon is observed in other generative architectures such as DCGAN and PGGAN. We further show that zero padding leads to an unbalanced spatial bias with a vague relation between locations. To offer a better spatial inductive bias, we investigate alternative positional encodings and analyze their effects. Based on a more flexible positional encoding explicitly, we propose a new multi-scale training strategy and demonstrate its effectiveness in the state-of-the-art unconditional generator StyleGAN2. Besides, the explicit spatial inductive bias substantially improve SinGAN for more versatile image manipulation.