ICCV 2021 论文列表

Editing raster text is a promising but challenging task. We propose to apply text vectorization for the task of raster text editing in display media, such as posters, web pages, or advertisements. In our approach, instead of applying image transformation or generation in the raster domain, we learn a text vectorization model to parse all the rendering parameters including text, location, size, font, style, effects, and hidden background, then utilize those parameters for reconstruction and any editing task. Our text vectorization takes advantage of differentiable text rendering to accurately reproduce the input raster text in a resolution-free parametric format. We show in the experiments that our approach can successfully parse text, styling, and background information in the unified model, and produces artifact-free text editing compared to a raster baseline.

Universal adversarial perturbation (UAP), i.e. a single perturbation to fool the network for most images, is widely recognized as a more practical attack because the UAP can be generated beforehand and applied directly during the attack stage. One intriguing phenomenon regarding untargeted UAP is that most images are misclassified to a dominant label. This phenomenon has been reported in previous works while lacking a justified explanation, for which our work attempts to provide an alternative explanation. For a more practical universal attack, our investigation of untargeted UAP focuses on alleviating the dependence on the original training samples, from removing the need for sample labels to limiting the sample size. Towards strictly data-free untargeted UAP, our work proposes to exploit artificial Jigsaw images as the training samples, demonstrating competitive performance. We further investigate the possibility of exploiting the UAP for a data-free black-box attack which is arguably the most practical yet challenging threat model. We demonstrate that there exists optimization-free repetitive patterns which can successfully attack deep models. Code is available at https://bit.ly/3y0ZTIC.

This paper proposes a novel weakly supervised approach for anomaly detection, which begins with a relation-aware feature extractor to capture the multi-scale convolutional neural network (CNN) features from a video. Afterwards, self-attention is integrated with conditional random fields (CRFs), the core of the network, to make use of the ability of self-attention in capturing the short-range correlations of the features and the ability of CRFs in learning the inter-dependencies of these features. Such a framework can learn not only the spatio-temporal interactions among the actors which are important for detecting complex movements, but also their short- and long-term dependencies across frames. Also, to deal with both local and non-local relationships of the features, a new variant of self-attention is developed by taking into consideration a set of cliques with different temporal localities. Moreover, a contrastive multi-instance learning scheme is considered to broaden the gap between the normal and abnormal instances, resulting in more accurate abnormal discrimination. Simulations reveal that the new method provides superior performance to the state-of-the-art works on the widespread UCF-Crime and ShanghaiTech datasets.

We propose a novel pointwise descriptor, called DWKS, aimed at finding correspondences across two deformable shape collections. Unlike the majority of existing descriptors, rather than capturing local geometry, DWKS captures the deformation around a point within a collection in a multi-scale and informative manner. This, in turn, allows to compute inter-collection correspondences without using landmarks. To this end, we build upon the successful spectral WKS descriptors, but rather than using the Laplace-Beltrami operator, show that a similar construction can be performed on shape difference operators, that capture differences or distortion within a collection. By leveraging the collection information our descriptor facilitates difficult non-rigid shape matching tasks, even in the presence of strong partiality and significant deformations. We demonstrate the utility of our approach across a range of challenging matching problems on both meshes and point clouds. The code for this paper can be found at https://github.com/RobinMagnet/DWKS.

Compared with traditional methods, the deep learning-based multi-focus image fusion methods can effectively improve the performance of image fusion tasks. However, the existing deep learning-based methods encounter a common issue of a large number of parameters, which leads to the deep learning models with high time complexity and low fusion efficiency. To address this issue, we propose a novel discrete Tchebichef moment-based Deep neural network, termed as DTMNet, for multi-focus image fusion. The proposed DTMNet is an end-to-end deep neural network with only one convolutional layer and three fully connected layers. The convolutional layer is fixed with DTM coefficients (DTMConv) to extract high/low-frequency information without learning parameters effectively. The three fully connected layers have learnable parameters for feature classification. Therefore, the proposed DTMNet for multi-focus image fusion has a small number of parameters (0.01M paras vs. 4.93M paras of regular CNN) and high computational efficiency (0.32s vs. 79.09s by regular CNN to fuse an image). In addition, a large-scale multi-focus image dataset is synthesized for training and verifying the deep learning model. Experimental results on three public datasets demonstrate that the proposed method is competitive with or even outperforms the state-of-the-art multi-focus image fusion methods in terms of subjective visual perception and objective evaluation metrics.

Traffic accident anticipation aims to accurately and promptly predict the occurrence of a future accident from dashcam videos, which is vital for a safety-guaranteed self-driving system. To encourage an early and accurate decision, existing approaches typically focus on capturing the cues of spatial and temporal context before a future accident occurs. However, their decision-making lacks visual explanation and ignores the dynamic interaction with the environment. In this paper, we propose Deep ReInforced accident anticipation with Visual Explanation, named DRIVE. The method simulates both the bottom-up and top-down visual attention mechanism in a dashcam observation environment so that the decision from the proposed stochastic multi-task agent can be visually explained by attentive regions. Moreover, the proposed dense anticipation reward and sparse fixation reward are effective in training the DRIVE model with our improved reinforcement learning algorithm. Experimental results show that the DRIVE model achieves state-of-the-art performance on multiple real-world traffic accident datasets. Code and pre-trained models are available at https://www.rit.edu/actionlab/drive.

We present a novel and flexible architecture for point cloud segmentation with dual-representation iterative learning. In point cloud processing, different representations have their own pros and cons. Thus, finding suitable ways to represent point cloud data structure while keeping its own internal physical property such as permutation and scale-invariant is a fundamental problem. Therefore, we propose our work, DRINet, which serves as the basic network structure for dual-representation learning with great flexibility at feature transferring and less computation cost, especially for large-scale point clouds. DRINet mainly consists of two modules called Sparse Point-Voxel Feature Extraction and Sparse Voxel-Point Feature Extraction. By utilizing these two modules iteratively, features can be propagated between two different representations. We further propose a novel multi-scale pooling layer for pointwise locality learning to improve context information propagation. Our network achieves state-of-the-art results for point cloud classification and segmentation tasks on several datasets while maintaining high runtime efficiency. For large-scale outdoor scenarios, our method outperforms state-of-the-art methods with a real-time inference speed of 62ms per frame.

In this work, we propose a Dynamic ResBlock Generative Adversarial Network (DRB-GAN) for artistic style transfer. The style code is modeled as the shared parameters for Dynamic ResBlocks connecting both the style encoding network and the style transfer network. In the style encoding network, a style class-aware attention mechanism is used to attend the style feature represent for generating the style codes. In the style transfer network, multiple Dynamic ResBlocks are designed to integrate the style code and the extracted CNN semantic feature and and then feed into the spatial window Layer-Instance Normalization (SW-LIN) decoder, which enables high-quality synthetic images with artistic style transfer. Moreover, the style collection conditional discriminator is designed to ensure our DRB-GAN model to equip with abilities for both arbitrary style transfer and collection style transfer during the training stage. No matter for arbitrary style transfer or collection style transfer, extensive experimental results strongly demonstrate that our proposed DRB-GAN beats state-of-the-art methods and exhibits its superior performance in terms of visual quality and efficiency.

Visual surface anomaly detection aims to detect local image regions that significantly deviate from normal appearance. Recent surface anomaly detection methods rely on generative models to accurately reconstruct the normal areas and to fail on anomalies. These methods are trained only on anomaly-free images, and often require hand-crafted post-processing steps to localize the anomalies, which prohibits optimizing the feature extraction for maximal detection capability. In addition to reconstructive approach, we cast surface anomaly detection primarily as a discriminative problem and propose a discriminatively trained reconstruction anomaly embedding model (DRAEM). The proposed method learns a joint representation of an anomalous image and its anomaly-free reconstruction, while simultaneously learning a decision boundary between normal and anomalous examples. The method enables direct anomaly localization without the need for additional complicated post-processing of the network output and can be trained using simple and general anomaly simulations. On the challenging MVTec anomaly detection dataset, DRAEM outperforms the current state-of-the-art unsupervised methods by a large margin and even delivers detection performance close to the fully-supervised methods on the widely used DAGM surface-defect detection dataset, while substantially outperforming them in localization accuracy.

Image Retrieval is a fundamental task of obtaining images similar to the query one from a database. A common image retrieval practice is to firstly retrieve candidate images via similarity search using global image features and then re-rank the candidates by leveraging their local features. Previous learning-based studies mainly focus on either global or local image representation learning to tackle the retrieval task. In this paper, we abandon the two-stage paradigm and seek to design an effective single-stage solution by integrating local and global information inside images into compact image representations. Specifically, we propose a Deep Orthogonal Local and Global (DOLG) information fusion framework for end-to-end image retrieval. It attentively extracts representative local information with multi-atrous convolutions and self-attention at first. Components orthogonal to the global image representation are then extracted from the local information. At last, the orthogonal components are concatenated with the global representation as a complementary, and then aggregation is performed to generate the final representation. The whole framework is end-to-end differentiable and can be trained with image-level labels. Extensive experimental results validate the effectiveness of our solution and show that our model achieves state-of-the-art image retrieval performances on Revisited Oxford and Paris datasets.

Human Pose Estimation (HPE) aims at retrieving the 3D position of human joints from images or videos. We show that current 3D HPE methods suffer a lack of viewpoint equivariance, namely they tend to fail or perform poorly when dealing with viewpoints unseen at training time. Deep learning methods often rely on either scale-invariant, translation-invariant, or rotation-invariant operations, such as max-pooling. However, the adoption of such procedures does not necessarily improve viewpoint generalization, rather leading to more data-dependent methods. To tackle this issue, we propose a novel capsule autoencoder network with fast Variational Bayes capsule routing, named DECA. By modeling each joint as a capsule entity, combined with the routing algorithm, our approach can preserve the joints' hierarchical and geometrical structure in the feature space, independently from the viewpoint. By achieving viewpoint equivariance, we drastically reduce the network data dependency at training time, resulting in an improved ability to generalize for unseen viewpoints. In the experimental validation, we outperform other methods on depth images from both seen and unseen viewpoints, both top-view, and front-view. In the RGB domain, the same network gives state-of-the-art results on the challenging viewpoint transfer task, also establishing a new framework for top-view HPE.

Spiking Neural Networks (SNNs) offer a promising alternative to traditional deep learning frameworks, since they provide higher computational efficiency due to event-driven information processing. SNNs distribute the analog values of pixel intensities into binary spikes over time. However, the most widely used input coding schemes, such as Poisson based rate-coding, do not leverage the additional temporal learning capability of SNNs effectively. Moreover, these SNNs suffer from high inference latency which is a major bottleneck to their deployment. To overcome this, we propose a time-based encoding scheme that utilizes the Discrete Cosine Transform (DCT) to reduce the number of timesteps required for inference. DCT decomposes an image into a weighted sum of sinusoidal basis images. At each time step, a single frequency base, taken in order and modulated by its corresponding DCT coefficient, is input to an accumulator that generates spikes upon crossing a threshold. We use the proposed scheme to learn DCT-SNN, a low-latency deep SNN with leaky-integrate-and-fire neurons, trained using surrogate gradient descent based backpropagation. We achieve top-1 accuracy of 89.94%, 68.3% and 52.43% on CIFAR-10, CIFAR-100 and TinyImageNet, respectively using VGG architectures. Notably, DCT-SNN performs inference with 2-14X reduced latency compared to other state-of-the-art SNNs, while achieving comparable accuracy to their standard deep learning counterparts. The dimension of the transform allows us to control the number of timesteps required for inference. Additionally, we can trade-off accuracy with latency in a principled manner by dropping the highest frequency components during inference. The code is publicly available*.

Shadow removal from a single image is generally still an open problem. Most existing learning-based methods use supervised learning and require a large number of paired images (shadow and corresponding non-shadow images) for training. A recent unsupervised method, Mask-ShadowGAN, addresses this limitation. However, it requires a binary mask to represent shadow regions, making it inapplicable to soft shadows. To address the problem, in this paper, we propose an unsupervised domain-classifier guided shadow removal network, DC-ShadowNet. Specifically, we propose to integrate a shadow/shadow-free domain classifier into a generator and its discriminator, enabling them to focus on shadow regions. To train our network, we introduce novel losses based on physics-based shadow-free chromaticity, shadow-robust perceptual features, and boundary smoothness. Moreover, we show that our network being unsupervised can be used for test-time training that further improves the results. Our experiments show that all these novel components allow our method to handle soft shadows, and also to perform better on hard shadows both quantitatively and qualitatively than the existing state-of-the-art shadow removal methods.

The field of face recognition (FR) has witnessed remarkable progress with the surge of deep learning. The effective loss functions play an important role for FR. In this paper, we observe that a majority of loss functions, including the widespread triplet loss and softmax-based cross-entropy loss, embed inter-class (negative) similarity s_n and intra-class (positive) similarity s_p into similarity pairs and optimize to reduce (s_n - s_p) in the training process. However, in the verification process, existing metrics directly take the absolute similarity between two features as the confidence of belonging to the same identity, which inevitably causes a gap between the training and verification process. To bridge the gap, we propose a new metric called Discrepancy Alignment Metric (DAM) for verification, which introduces the Local Inter-class Discrepancy (LID) for each face image to normalize the absolute similarity score. To estimate the LID of each face image in the verification process, we propose two types of LID Estimation (LIDE) methods, which are reference-based and learning-based estimation methods, respectively. The proposed DAM is plug-and-play and can be easily applied to the most existing methods. Extensive experiments on multiple popular face recognition benchmark datasets demonstrate the effectiveness of our proposed method.

Text-to-image synthesis refers to generating an image from a given text description, the key goal of which lies in photo realism and semantic consistency. Previous methods usually generate an initial image with sentence embedding and then refine it with fine-grained word embedding. Despite the significant progress, the 'aspect' information (e.g., red eyes) contained in the text, referring to several words rather than a word that depicts 'a particular part or feature of something', is often ignored, which is highly helpful for synthesizing image details. How to make better utilization of aspect information in text-to-image synthesis still remains an unresolved challenge. To address this problem, in this paper, we propose a Dynamic Aspect-awarE GAN (DAE-GAN) that represents text information comprehensively from multiple granularities, including sentence-level, word-level, and aspect-level. Moreover, inspired by human learning behaviors, we develop a novel Aspect-aware Dynamic Re-drawer (ADR) for image refinement, in which an Attended Global Refinement (AGR) module and an Aspect-aware Local Refinement (ALR) module are alternately employed. AGR utilizes word-level embedding to globally enhance the previously generated image, while ALR dynamically employs aspect-level embedding to refine image details from a local perspective. Finally, a corresponding matching loss function is designed to ensure the text-image semantic consistency at different levels. Extensive experiments on two well-studied and publicly available datasets (i.e., CUB-200 and COCO) demonstrate the superiority and rationality of our method.

This work proposes a weakly-supervised temporal action localization framework, called D2-Net, which strives to temporally localize actions using video-level supervision. Our main contribution is the introduction of a novel loss formulation, which jointly enhances the discriminability of latent embeddings and robustness of the output temporal class activations with respect to foreground-background noise caused by weak supervision. The proposed formulation comprises a discriminative and a denoising loss term for enhancing temporal action localization. The discriminative term incorporates a classification loss and utilizes a top-down attention mechanism to enhance the separability of latent foreground-background embeddings. The denoising loss term explicitly addresses the foreground-background noise in class activations by simultaneously maximizing intra-video and inter-video mutual information using a bottom-up attention mechanism. As a result, activations in the foreground regions are emphasized whereas those in the background regions are suppressed, thereby leading to more robust predictions. Comprehensive experiments are performed on multiple benchmarks, including THUMOS14 and ActivityNet1.2. Our D2-Net performs favorably in comparison to the existing methods on all datasets, achieving gains as high as 2.3% in terms of mAP at IoU=0.5 on THUMOS14. Source code is available at https://github.com/naraysa/D2-Net.

We present in this paper a new architecture, named Convolutional vision Transformer (CvT), that improves Vision Transformer (ViT) in performance and efficiency by introducing convolutions into ViT to yield the best of both designs. This is accomplished through two primary modifications: a hierarchy of Transformers containing a new convolutional token embedding, and a convolutional Trasnsformer block leveraging a convolutional projection. These changes introduce desirable properties of convolutional neural networks (CNNs) to the ViT architecture (i.e. shift, scale, and distortion invariance) while maintaining the merits of Transformers (i.e. dynamic attention, global context, and better generalization). We validate CvT by conducting extensive experiments, showing that this approach achieves state-of-the-art performance over other Vision Transformers and ResNets on ImageNet-1k, with less parameters and lower FLOPs. In addition, performance gains are maintained when pretrained on larger datasets (e.g. ImageNet-22k) and fine-tuned to downstream tasks. Finally, our results show that the positional encoding, a crucial component in existing Vision Transformers, can be safely removed in our model, simplifying the design for higher resolution vision tasks. Code will be released at https://github.com/microsoft/CvT.

Few-shot learning describes the challenging problem of recognizing samples from unseen classes given very few labeled examples. In many cases, few-shot learning is cast as learning an embedding space that assigns test samples to their corresponding class prototypes. Previous methods assume that data of all few-shot learning tasks comply with a fixed geometrical structure, mostly a Euclidean structure. Questioning this assumption that is clearly difficult to hold in real-world scenarios and incurs distortions to data, we propose to learn a task-aware curved embedding space by making use of the hyperbolic geometry. As a result, task-specific embedding spaces where suitable curvatures are generated to match the characteristics of data are constructed, leading to more generic embedding spaces. We then leverage on intra-class and inter-class context information in the embedding space to generate class prototypes for discriminative classification. We conduct a comprehensive set of experiments on inductive and transductive few-shot learning, demonstrating the benefits of our proposed method over existing embedding methods.

Self-supervised visual representation learning has achieved remarkable success in recent years. By subverting the need for supervised labels, such approaches are able to utilize the numerous unlabeled images that exist on the Internet and in photographic datasets. Yet to build truly intelligent agents, we must construct representation learning algorithms that can learn not only from datasets but also learn in environments. An agent in a natural environment will not typically be fed curated data. Instead, it must explore its environment to acquire the data it will learn from. We propose a framework, curious representation learning (CRL), which jointly learns a reinforcement learning policy and a visual representation model. The policy is trained to maximize the error of the representation learner, and in doing so is incentivized to explore its environment. At the same time, the learned representation becomes stronger and stronger as the policy feeds it ever harder data to learn from. Our learned embodied representations enable promising transfer to downstream embodied semantic and language-guided navigation, performing better or comparable to ImageNet pretraining without using any supervision at all. In addition, despite being trained in simulation, our learned representations can obtain interpretable results on real images.