AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Differentiable Learning-to-Group Channels via Groupable Convolutional Neural Networks

Zhaoyang Zhang, Jingyu Li, Wenqi Shao, Zhanglin Peng, Ruimao Zhang, Xiaogang Wang, Ping Luo

ICCV 2019 · Architectures and efficiency › CNN design · approach: Compact CNN architectures · 41 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
learn group numbers per layer with DGConv instead of fixed 32 groups (b=32)ResNeXt50 · ImageNet valtop-1 accuracy: 77.9 -> 78.4Improved
learn group numbers per layer with DGConv instead of fixed 32 groups (b=32)ResNeXt101 · ImageNet valtop-1 accuracy: 79.1 -> 79.9Improved
increase scale constant b from 32 to 128 (less group-conv complexity) (ablation)G-ResNeXt50 · ImageNet valtop-1 accuracy: 78.4 -> 77.8Hurt
increase scale constant b from 32 to 256 (less group-conv complexity) (ablation)G-ResNeXt101 · ImageNet valtop-1 accuracy: 79.9 -> 79.0Hurt
reduce group conv FLOPs from 0.70x to 0.47x via b=64 (ablation)G-ResNeXt101 · ImageNet valtop-1 accuracy: 79.9 -> 79.8No clear effect
replace uniform groups with learned DGConv groups at b=96 (0.27x FLOPs)ResNeXt50 · ImageNet valtop-1 accuracy: 77.9 -> 78.0No clear effect
use DGConv at 2x ResNet101 complexity (b=2) instead of ResNeXt101 64x4dResNeXt101 · ImageNet valtop-1 accuracy: 79.8 -> 80.1Improved
re-train G-ResNeXt101 with a different random seedG-ResNeXt101 (b=32) · ImageNet valtop-1 accuracy: 79.9 -> 79.8No clear effect
fix ResNeXt50 group numbers to those learned by DGConv (b=32) and train a plain network (ablation)ResNeXt50 · ImageNet valtop-1 accuracy: 77.9 -> 78.3Improved
fix ResNeXt101 group numbers to those learned by DGConv (b=32) and train a plain network (ablation)ResNeXt101 · ImageNet valtop-1 accuracy: 79.1 -> 79.8Improved

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