AI Paper Atlas

Research map / Detection and segmentation / CNN features

CycleMLP: A MLP-like Architecture for Dense Prediction

Shoufa Chen, Enze Xie, Chongjian GE, Runjian Chen, Ding Liang, Ping Luo

ICLR 2022 · Detection and segmentation › CNN features · approach: Convolution-transformer backbones · 98 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
Replace a 7×1 branch with a 1×7 branch to retain both directional branches (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 80.4 -> 81.6Improved
Add the 1×1 branch to the 1×7 and 7×1 branches (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.3 -> 81.6Improved
Replace the 7×1 branch with a second 1×7 branch (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.6 -> 80.6Hurt
Replace the 1×7 branch with a second 7×1 branch (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.6 -> 80.5Hurt
Use stepsize 3 instead of 7 (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.6 -> 81.6No clear effect
Use stepsize 3 instead of 7 (ablation)CycleMLP-B2 · ADE20K valmIoU: 43.9 -> 42.4Hurt
Use stepsize 5 instead of 7 (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.6 -> 81.6No clear effect
Use stepsize 5 instead of 7 (ablation)CycleMLP-B2 · ADE20K valmIoU: 43.9 -> 43.2Hurt
Use stepsize 9 instead of 7 (ablation)CycleMLP-B2 · ImageNet-1K valTop-1 accuracy: 81.6 -> 81.5Hurt
Use stepsize 9 instead of 7 (ablation)CycleMLP-B2 · ADE20K valmIoU: 43.9 -> 43.2Hurt
Use cyclical rather than random sampling (ablation)CycleMLP · ImageNet-1KTop-1 accuracy: +1.3%Improved
Use dilated rather than random sampling (ablation)CycleMLP · ImageNet-1KTop-1 accuracy: +1.0% accImproved

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