AI Paper Atlas

Research map / Architectures and efficiency / Attention and transformers

Demystify Transformers & Convolutions in Modern Image Deep Networks

Xiaowei Hu, Min Yong Shi, Wei‐Yun Wang, Sitong Wu, Linjie Xing, Wenhai Wang, Xizhou Zhu, Lewei Lu, Jie Zhou, Xiaogang Wang, Yu Qiao, Jifeng Dai

TPAMI 2024 · Architectures and efficiency › Attention and transformers · approach: Efficient and robust vision transformers · 10 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
use InternImage deformable convolution instead of PVT spatial-reduction attentionunified architecture (Tiny) · ImageNet-Aaccuracy: 26.1 -> 31.9Improved
use deformable convolution v3 instead of depth-wise convolutionunified architecture (Tiny) · ImageNet-Aaccuracy: 23.3 -> 31.9Improved
use halo attention instead of PVT spatial-reduction attentionunified architecture (Tiny) · ImageNet-C1-mCE: 56.1 -> 57.9Improved
use PVT global attention instead of InternImage DCNv3unified architecture (Tiny) · FGSM eps=8/255adversarial accuracy: 41.2 -> 41.7No clear effect
use Swin shifted-window attention instead of halo attentionunified architecture (Tiny) · FGSM eps=8/255adversarial accuracy: 38 -> 32.7Hurt
use deformable convolution instead of PVT for translation invariance at 64 pxunified architecture (Tiny) · ImageNet-1k translationconsistency: 87.6 -> 89.2Improved
use deformable convolution instead of depth-wise convolution for rotation 45 degreesunified architecture (Tiny) · ImageNet-1k rotationaccuracy: 75.7 -> 78.3Improved
use deformable convolution instead of Swin for scale 0.25 on COCOunified architecture · COCObox mAP: 23.3 -> 25.2Improved
scale up model from Micro to TinyU-InternImage · ImageNet-1kclean accuracy: 75.3 -> 83.3Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy