arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2204.00993cs.CV
收藏

通过重新审视高频元件改进视觉Transformer

Improving Vision Transformers by Revisiting High-frequency Components

Peng Cheng Laboratory(鹏城实验室);Tsinghua Shenzhen International Graduate School, Tsinghua University(清华大学深圳国际研究生院);School of ECE at Peking University(北京大学电子工程与计算机科学学院);Tencent(腾讯);Sea AI Lab(Sea人工智能实验室)

Jiawang Bai, Li Yuan, Shu-Tao Xia, Shuicheng Yan, Zhifeng Li, Wei Liu

摘要:

Transformer模型在处理各种视觉任务中表现出了很好的效果。然而,与训练卷积神经网络(CNN)模型相比,训练视觉Transformer(ViT)模型更加困难,并且依赖于大规模的训练集。为了解释这一现象,我们提出了一个假设,即\textit{ViT models are less effective in capturing the high-frequency components of images than CNN models},并验证了它的频率分析。受这一发现的启发,我们首先从新的频率角度研究了现有技术对改善ViT模型的影响,并发现一些技术(例如,RandAugment)可以归因于更好地使用高频分量。然后,为了弥补ViT模型能力的不足,我们提出了HAT,它通过对抗训练直接增强图像的高频分量。我们表明,HAT可以始终如一地提高各种ViT模型的性能(例如,ViT-B +1.2%,Swin-B +0.5%),特别是将仅使用ImageNet-1 K数据的高级模型VOLO-D5提升至87.3%,并且在分发外数据上也可以保持优势,并转移到下游任务。该代码可从以下网址获取:https://github.com/jiawangbai/HAT。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑