AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Drop an Octave: Reducing Spatial Redundancy in Convolutional Neural Networks With Octave Convolution

Yunpeng Chen, Haoqi Fan, Bing Xu, Zhicheng Yan, Yannis Kalantidis, Marcus Rohrbach, Shuicheng Yan, Jiashi Feng

ICCV 2019 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 800 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 convolutions with OctConv at alpha=0.125 (ablation)ResNet-50 · ImageNet valtop-1 accuracy: 77.0 -> 78.2Improved
raise low-frequency ratio alpha to 0.75 (ablation)OctConv ResNet-50 · ImageNet valtop-1 accuracy: 77.0 -> 76.7 (4.1 -> 1.9 GFLOPs)Mixed
add Low->High information path (ablation)OctConv ResNet-50 alpha 0.5 · ImageNet valtop-1 accuracy: 76.0 -> 76.4Improved
add High->Low information path (ablation)OctConv ResNet-50 alpha 0.5 · ImageNet valtop-1 accuracy: 76.0 -> 76.4Improved
use average pooling instead of strided convolution for down-sampling (ablation)OctConv ResNet-50 alpha 0.5 · ImageNet valtop-1 accuracy: 76.3 -> 77.4Improved
use average pooling instead of max pooling for down-sampling (ablation)OctConv ResNet-50 alpha 0.5 · ImageNet valtop-1 accuracy: 77.0 -> 77.4Improved
larger test resolution (256 to 740) widens OctConv gain over baseline (ablation)OctConv ResNet-50 alpha 0.5 · ImageNet valtop-1 accuracy: +0.7 -> +1.2Improved
apply OctConv to shallow ResNet-26 (alpha 0.25) (ablation)ResNet-26 · ImageNet valtop-1 accuracy: 73.2 -> 76.1Improved
use OctConv with wider model at same FLOPsMobileNet v1 · ImageNet valtop-1 accuracy: 70.3 -> 72.5Improved
apply OctConv alpha=0.1I3D trained from scratch · Kinetics-400top-1 accuracy: 72.6 -> 73.6Improved
apply OctConv alpha=0.5 (ablation)I3D trained from scratch · Kinetics-400top-1 accuracy: 72.6 -> 72.1Hurt
apply OctConv alpha=0.1C2D ImageNet pretrained · Kinetics-400top-1 accuracy: 71.9 -> 73.8Improved

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