This paper presents a new vision Transformer, named Iwin Transformer, which is specifically designed for human-object interaction (HOI) detection, a detailed scene understanding task involving a sequential process of human/object detection and interaction recognition. Iwin Transformer is a hierarchical Transformer which progressively performs token representation learning and token agglomeration within irregular windows. The irregular windows, achieved by augmenting regular grid locations with learned offsets, 1) eliminate redundancy in token representation learning, which leads to efficient human/object detection, and 2) enable the agglomerated tokens to align with humans/objects with different shapes, which facilitates the acquisition of highly-abstracted visual semantics for interaction recognition. The effectiveness and efficiency of Iwin Transformer are verified on the two standard HOI detection benchmark datasets, HICO-DET and V-COCO. Results show our method outperforms existing Transformers-based methods by large margins (3.7 mAP gain on HICO-DET and 2.0 mAP gain on V-COCO) with fewer training epochs ($0.5 \times$).
Research on the generalization ability of deep neural networks (DNNs) has recently attracted a great deal of attention. However, due to their complex architectures and large numbers of parameters, measuring the generalization ability of specific DNN models remains an open challenge. In this paper, we propose to use multiple factors to measure and rank the relative generalization of DNNs based on a new concept of confidence dimension (CD). Furthermore, we provide a feasible framework in our CD to theoretically calculate the upper bound of generalization based on the conventional Vapnik-Chervonenk dimension (VC-dimension) and Hoeffding's inequality. Experimental results on image classification and object detection demonstrate that our CD can reflect the relative generalization ability for different DNNs. In addition to full-precision DNNs, we also analyze the generalization ability of binary neural networks (BNNs), whose generalization ability remains an unsolved problem. Our CD yields a consistent and reliable measure and ranking for both full-precision DNNs and BNNs on all the tasks.
Deep learning has been regarded as the `go to' solution for many tasks today, but its intrinsic vulnerability to malicious attacks has become a major concern. The vulnerability is affected by a variety of factors including models, tasks, data, and attackers. Consequently, methods such as Adversarial Training and Randomized Smoothing have been proposed to tackle the problem in a wide range of applications. In this paper, we investigate skeleton-based Human Activity Recognition, which is an important type of time-series data but under-explored in defense against attacks. Our method is featured by (1) a new Bayesian Energy-based formulation of robust discriminative classifiers, (2) a new parameterization of the adversarial sample manifold of actions, and (3) a new post-train Bayesian treatment on both the adversarial samples and the classifier. We name our framework Bayesian Energy-based Adversarial Training or BEAT. BEAT is straightforward but elegant, which turns vulnerable black-box classifiers into robust ones without sacrificing accuracy. It demonstrates surprising and universal effectiveness across a wide range of action classifiers and datasets, under various attacks.
Recently, Transformers have shown promising performance in various vision tasks. To reduce the quadratic computation complexity caused by each query attending to all keys/values, various methods have constrained the range of attention within local regions, where each query only attends to keys/values within a hand-crafted window. However, these hand-crafted window partition mechanisms are data-agnostic and ignore their input content, so it is likely that one query maybe attends to irrelevant keys/values. To address this issue, we propose a Dynamic Group Attention (DG-Attention), which dynamically divides all queries into multiple groups and selects the most relevant keys/values for each group. Our DG-Attention can flexibly model more relevant dependencies without any spatial constraint that is used in hand-crafted window based attention. Built on the DG-Attention, we develop a general vision transformer backbone named Dynamic Group Transformer (DGT). Extensive experiments show that our models can outperform the state-of-the-art methods on multiple common vision tasks, including image classification, semantic segmentation, object detection, and instance segmentation.
A human's attention can intuitively adapt to corrupted areas of an image by recalling a similar uncorrupted image they have previously seen. This observation motivates us to improve the attention of adversarial images by considering their clean counterparts. To accomplish this, we introduce Associative Adversarial Learning (AAL) into adversarial learning to guide a selective attack. We formulate the intrinsic relationship between attention and attack (perturbation) as a coupling optimization problem to improve their interaction. This leads to an attention backtracking algorithm that can effectively enhance the attention's adversarial robustness. Our method is generic and can be used to address a variety of tasks by simply choosing different kernels for the associative attention that select other regions for a specific attack. Experimental results show that the selective attack improves the model's performance. We show that our method improves the recognition accuracy of adversarial training on ImageNet by 8.32% compared with the baseline. It also increases object detection mAP on PascalVOC by 2.02% and recognition accuracy of few-shot learning on miniImageNet by 1.63%.
Recently, Transformers have shown promising performance in various vision tasks. To reduce the quadratic computation complexity caused by the global self-attention, various methods constrain the range of attention within a local region to improve its efficiency. Consequently, their receptive fields in a single attention layer are not large enough, resulting in insufficient context modeling. To address this issue, we propose a Pale-Shaped self-Attention (PS-Attention), which performs self-attention within a pale-shaped region. Compared to the global self-attention, PS-Attention can reduce the computation and memory costs significantly. Meanwhile, it can capture richer contextual information under the similar computation complexity with previous local self-attention mechanisms. Based on the PS-Attention, we develop a general Vision Transformer backbone with a hierarchical architecture, named Pale Transformer, which achieves 83.4%, 84.3%, and 84.9% Top-1 accuracy with the model size of 22M, 48M, and 85M respectively for 224 ImageNet-1K classification, outperforming the previous Vision Transformer backbones. For downstream tasks, our Pale Transformer backbone performs better than the recent state-of-the-art CSWin Transformer by a large margin on ADE20K semantic segmentation and COCO object detection & instance segmentation. The code will be released on https://github.com/BR-IDL/PaddleViT.
Real-time point cloud processing is fundamental for lots of computer vision tasks, while still challenged by the computational problem on resource-limited edge devices. To address this issue, we implement XNOR-Net-based binary neural networks (BNNs) for an efficient point cloud processing, but its performance is severely suffered due to two main drawbacks, Gaussian-distributed weights and non-learnable scale factor. In this paper, we introduce point-wise operations based on Expectation-Maximization (POEM) into BNNs for efficient point cloud processing. The EM algorithm can efficiently constrain weights for a robust bi-modal distribution. We lead a well-designed reconstruction loss to calculate learnable scale factors to enhance the representation capacity of 1-bit fully-connected (Bi-FC) layers. Extensive experiments demonstrate that our POEM surpasses existing the state-of-the-art binary point cloud networks by a significant margin, up to 6.7 %.
Facial age estimation is an important yet very challenging problem in computer vision. To improve the performance of facial age estimation, we first formulate a simple standard baseline and build a much strong one by collecting the tricks in pre-training, data augmentation, model architecture, and so on. Compared with the standard baseline, the proposed one significantly decreases the estimation errors. Moreover, long-tailed recognition has been an important topic in facial age datasets, where the samples often lack on the elderly and children. To train a balanced age estimator, we propose a two-stage training method named Long-tailed Age Estimation (LAE), which decouples the learning procedure into representation learning and classification. The effectiveness of our approach has been demonstrated on the dataset provided by organizers of Guess The Age Contest 2021.
Face image animation from a single image has achieved remarkable progress. However, it remains challenging when only sparse landmarks are available as the driving signal. Given a source face image and a sequence of sparse face landmarks, our goal is to generate a video of the face imitating the motion of landmarks. We develop an efficient and effective method for motion transfer from sparse landmarks to the face image. We then combine global and local motion estimation in a unified model to faithfully transfer the motion. The model can learn to segment the moving foreground from the background and generate not only global motion, such as rotation and translation of the face, but also subtle local motion such as the gaze change. We further improve face landmark detection on videos. With temporally better aligned landmark sequences for training, our method can generate temporally coherent videos with higher visual quality. Experiments suggest we achieve results comparable to the state-of-the-art image driven method on the same identity testing and better results on cross identity testing.