2022-07-29 更新
Improving Vision Transformers by Revisiting High-frequency Components
Authors:Jiawang Bai, Li Yuan, Shu-Tao Xia, Shuicheng Yan, Zhifeng Li, Wei Liu
The transformer models have shown promising effectiveness in dealing with various vision tasks. However, compared with training Convolutional Neural Network (CNN) models, training Vision Transformer (ViT) models is more difficult and relies on the large-scale training set. To explain this observation we make a hypothesis that \textit{ViT models are less effective in capturing the high-frequency components of images than CNN models}, and verify it by a frequency analysis. Inspired by this finding, we first investigate the effects of existing techniques for improving ViT models from a new frequency perspective, and find that the success of some techniques (e.g., RandAugment) can be attributed to the better usage of the high-frequency components. Then, to compensate for this insufficient ability of ViT models, we propose HAT, which directly augments high-frequency components of images via adversarial training. We show that HAT can consistently boost the performance of various ViT models (e.g., +1.2% for ViT-B, +0.5% for Swin-B), and especially enhance the advanced model VOLO-D5 to 87.3% that only uses ImageNet-1K data, and the superiority can also be maintained on out-of-distribution data and transferred to downstream tasks. The code is available at: https://github.com/jiawangbai/HAT.
PDF Accepted to ECCV2022; Code: https://github.com/jiawangbai/HAT
点此查看论文截图
DnSwin: Toward Real-World Denoising via Continuous Wavelet Sliding-Transformer
Authors:Hao Li, Zhijing Yang, Xiaobin Hong, Ziying Zhao, Junyang Chen, Yukai Shi, Jinshan Pan
Real-world image denoising is a practical image restoration problem that aims to obtain clean images from in-the-wild noisy input. Recently, Vision Transformer (ViT) exhibits a strong ability to capture long-range dependencies and many researchers attempt to apply ViT to image denoising tasks. However, real-world image is an isolated frame that makes the ViT build the long-range dependencies on the internal patches, which divides images into patches and disarranges the noise pattern and gradient continuity. In this article, we propose to resolve this issue by using a continuous Wavelet Sliding-Transformer that builds frequency correspondence under real-world scenes, called DnSwin. Specifically, we first extract the bottom features from noisy input images by using a CNN encoder. The key to DnSwin is to separate high-frequency and low-frequency information from the features and build frequency dependencies. To this end, we propose Wavelet Sliding-Window Transformer that utilizes discrete wavelet transform, self-attention and inverse discrete wavelet transform to extract deep features. Finally, we reconstruct the deep features into denoised images using a CNN decoder. Both quantitative and qualitative evaluations on real-world denoising benchmarks demonstrate that the proposed DnSwin performs favorably against the state-of-the-art methods.
PDF DnSwin, a continuous Wavelet Sliding-Transformer, builds frequency correspondence under real-world scenes for image denoising
点此查看论文截图
Self-slimmed Vision Transformer
Authors:Zhuofan Zong, Kunchang Li, Guanglu Song, Yali Wang, Yu Qiao, Biao Leng, Yu Liu
Vision transformers (ViTs) have become the popular structures and outperformed convolutional neural networks (CNNs) on various vision tasks. However, such powerful transformers bring a huge computation burden, because of the exhausting token-to-token comparison. The previous works focus on dropping insignificant tokens to reduce the computational cost of ViTs. But when the dropping ratio increases, this hard manner will inevitably discard the vital tokens, which limits its efficiency. To solve the issue, we propose a generic self-slimmed learning approach for vanilla ViTs, namely SiT. Specifically, we first design a novel Token Slimming Module (TSM), which can boost the inference efficiency of ViTs by dynamic token aggregation. As a general method of token hard dropping, our TSM softly integrates redundant tokens into fewer informative ones. It can dynamically zoom visual attention without cutting off discriminative token relations in the images, even with a high slimming ratio. Furthermore, we introduce a concise Feature Recalibration Distillation (FRD) framework, wherein we design a reverse version of TSM (RTSM) to recalibrate the unstructured token in a flexible auto-encoder manner. Due to the similar structure between teacher and student, our FRD can effectively leverage structure knowledge for better convergence. Finally, we conduct extensive experiments to evaluate our SiT. It demonstrates that our method can speed up ViTs by 1.7x with negligible accuracy drop, and even speed up ViTs by 3.6x while maintaining 97% of their performance. Surprisingly, by simply arming LV-ViT with our SiT, we achieve new state-of-the-art performance on ImageNet. Code is available at https://github.com/Sense-X/SiT.
PDF Accepted by ECCV 2022. Code is available at https://github.com/Sense-X/SiT
点此查看论文截图