AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Revisiting Dynamic Convolution via Matrix Decomposition

Yunsheng Li, Yinpeng Chen, Xiyang Dai, mengchen liu, Dongdong Chen, Ye Yu, Lu Yuan, Zicheng Liu, Mei Chen, Nuno Vasconcelos

ICLR 2021 · Architectures and efficiency › CNN design · approach: Large and dynamic kernel design · 12 citations (OpenAlex)

Read the paper

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 dynamic attention over kernel groups with low-dimensional dynamic channel fusionMobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 65.4 -> 70.2Improved
Add dynamic channel-wise attention to the static kernel (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 69.2 -> 69.8Improved
Add dynamic channel fusion to the static kernel (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 65.4 -> 69.2Improved
Add dynamic channel-wise attention to the static kernel (ablation)DCD on ResNet-18 · ImageNetTop-1 accuracy: 70.4 -> 71.5Improved
Add dynamic channel fusion to the static kernel (ablation)DCD on ResNet-18 · ImageNetTop-1 accuracy: 70.4 -> 72.8Improved
Reduce latent dimension L to ×0.25 of default (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 69.8 -> 68.7Hurt
Reduce latent dimension L to ×0.50 of default (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 69.8 -> 69.0Hurt
Reduce latent dimension L to ×0.75 of default (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 69.8 -> 69.6Hurt
Use a block-diagonal dynamic residual with B=8 (ablation)DCD on MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 69.8 -> 68.5Hurt
Apply DCD to pointwise convolution and classifier rather than all three layer types (ablation)MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 70.0 -> 70.2Improved
Add DCD to depthwise convolution when pointwise convolution and classifier already use DCD (ablation)MobileNetV2 ×0.5 · ImageNetTop-1 accuracy: 70.2 -> 70.0Hurt
Use channel fusion alone instead of joint channel and kernel-element fusion (ablation)DCD on ResNet-18 3×3 convolutions · ImageNetTop-1 accuracy: 71.3 -> 73.1Improved

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