AI Paper Atlas

Research map / Architectures and efficiency / CNN design

Aggregated Residual Transformations for Deep Neural Networks

Saining Xie, Ross Girshick, Piotr Dollar, Zhuowen Tu, Kaiming He

CVPR 2017 · Architectures and efficiency › CNN design · approach: Efficient CNN scaling and dynamic inference · 12,036 citations (OpenAlex)

Check 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 the ResNet-50 bottleneck with aggregated transformations at cardinality 32, keeping complexity similar (ablation)ResNet-50 · ImageNet-1K valtop-1 error: 23.9 -> 22.2Improved
increase cardinality while reducing bottleneck width to preserve complexity (ablation)ResNet-50 · ImageNet-1K valtop-1 error: 23.9 -> 23.0 -> 22.6 -> 22.3 -> 22.2Improved
increase cardinality while reducing bottleneck width to preserve complexity (ablation)ResNet-101 · ImageNet-1K valtop-1 error: 22.0 -> 21.7 -> 21.4 -> 21.3 -> 21.2Improved
increase cardinality to 64 at approximately 2× complexity (ablation)ResNeXt-101 (32×4d) · ImageNet-1K valtop-1 error: 21.2 -> 20.4Improved
spend approximately 2× complexity on increasing cardinality rather than depth or width (ablation)ResNet-101 · ImageNet-1K valtop-1 error: ResNet-200: 21.7; wider ResNet-101: 21.3; ResNeXt-101 (2×64d): 20.7Improved
remove residual shortcuts (ablation)ResNeXt-50 (32×4d) · ImageNet-1K valtop-1 error: 22.2 -> 26.1Hurt
remove residual shortcuts (ablation)ResNet-50 · ImageNet-1K valtop-1 error: 23.9 -> 31.2Hurt
replace ResNet with ResNeXt-50 (32×4d) at similar complexityResNet-50 · ImageNet-5K train, 5K-way evaluationtop-1 error: 45.5 -> 42.3Improved
replace ResNet with ResNeXt-101 (32×4d) at similar complexityResNet-101 · ImageNet-5K train, 5K-way evaluationtop-1 error: 42.4 -> 40.1Improved
use cardinality 16 rather than a single wide residual pathCIFAR-10 and CIFAR-100 models · CIFAR test setstest error: CIFAR-10: 4.17 -> 3.58; CIFAR-100: 20.50 -> 17.31Improved
replace ResNet-50 with ResNeXt-50 (32×4d) at similar complexityFaster R-CNN with ResNet-50 · COCO minivalAP: 26.5 -> 27.5Improved
replace ResNet-101 with ResNeXt-101 (32×4d) at similar complexityFaster R-CNN with ResNet-101 · COCO minivalAP: 29.8 -> 30.0Improved

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