Demystify Transformers & Convolutions in Modern Image Deep Networks

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Hu, Xiaowei, Shi, Min, Wang, Weiyun, Wu, Sitong, Xing, Linjie, Wang, Wenhai, Zhu, Xizhou, Lu, Lewei, Zhou, Jie, Wang, Xiaogang, Qiao, Yu, Dai, Jifeng
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866911012545888256
author Hu, Xiaowei
Shi, Min
Wang, Weiyun
Wu, Sitong
Xing, Linjie
Wang, Wenhai
Zhu, Xizhou
Lu, Lewei
Zhou, Jie
Wang, Xiaogang
Qiao, Yu
Dai, Jifeng
author_facet Hu, Xiaowei
Shi, Min
Wang, Weiyun
Wu, Sitong
Xing, Linjie
Wang, Wenhai
Zhu, Xizhou
Lu, Lewei
Zhou, Jie
Wang, Xiaogang
Qiao, Yu
Dai, Jifeng
contents Vision transformers have gained popularity recently, leading to the development of new vision backbones with improved features and consistent performance gains. However, these advancements are not solely attributable to novel feature transformation designs; certain benefits also arise from advanced network-level and block-level architectures. This paper aims to identify the real gains of popular convolution and attention operators through a detailed study. We find that the key difference among these feature transformation modules, such as attention or convolution, lies in their spatial feature aggregation approach, known as the "spatial token mixer" (STM). To facilitate an impartial comparison, we introduce a unified architecture to neutralize the impact of divergent network-level and block-level designs. Subsequently, various STMs are integrated into this unified framework for comprehensive comparative analysis. Our experiments on various tasks and an analysis of inductive bias show a significant performance boost due to advanced network-level and block-level designs, but performance differences persist among different STMs. Our detailed analysis also reveals various findings about different STMs, including effective receptive fields, invariance, and adversarial robustness tests.
format Preprint
id arxiv_https___arxiv_org_abs_2211_05781
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle Demystify Transformers & Convolutions in Modern Image Deep Networks
Hu, Xiaowei
Shi, Min
Wang, Weiyun
Wu, Sitong
Xing, Linjie
Wang, Wenhai
Zhu, Xizhou
Lu, Lewei
Zhou, Jie
Wang, Xiaogang
Qiao, Yu
Dai, Jifeng
Computer Vision and Pattern Recognition
Vision transformers have gained popularity recently, leading to the development of new vision backbones with improved features and consistent performance gains. However, these advancements are not solely attributable to novel feature transformation designs; certain benefits also arise from advanced network-level and block-level architectures. This paper aims to identify the real gains of popular convolution and attention operators through a detailed study. We find that the key difference among these feature transformation modules, such as attention or convolution, lies in their spatial feature aggregation approach, known as the "spatial token mixer" (STM). To facilitate an impartial comparison, we introduce a unified architecture to neutralize the impact of divergent network-level and block-level designs. Subsequently, various STMs are integrated into this unified framework for comprehensive comparative analysis. Our experiments on various tasks and an analysis of inductive bias show a significant performance boost due to advanced network-level and block-level designs, but performance differences persist among different STMs. Our detailed analysis also reveals various findings about different STMs, including effective receptive fields, invariance, and adversarial robustness tests.
title Demystify Transformers & Convolutions in Modern Image Deep Networks
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2211.05781