AI Paper Atlas

Research map / Label-efficient and robust learning / Test-time adaptation

Better Aggregation in Test-Time Augmentation

Divya Shanmugam, Davis Blalock, Guha Balakrishnan, John Guttag

ICCV 2021 · Label-efficient and robust learning › Test-time adaptation · approach: Test-time adaptation of vision-language models · 280 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 mean with learned per-augmentation/class weightsTTA aggregation, MobileNetV2 · Flowers102 (standard policy)Top-1 accuracy: 90.47 -> 92.62Improved
replace mean with learned weightsTTA aggregation, InceptionV3 · ImageNet (standard policy)Top-1 accuracy: 72.45 -> 72.79Improved
replace mean with learned weightsTTA aggregation, MobileNetV2 · ImageNet (standard policy)Top-1 accuracy: 72.69 -> 72.43Hurt
replace mean with learned weightsTTA aggregation, CNN-7 · CIFAR100 (expanded policy)Top-1 accuracy: 73.41 -> 73.18Hurt
replace mean with learned weightsTTA aggregation, CNN-5 · STL10 (standard policy)Top-1 accuracy: 78.58 -> 78.52No clear effect
replace mean with learned weightsTTA aggregation, ResNet-50 · Flowers102 (expanded policy)Top-1 accuracy: 92.48 -> 93.29Improved
use mean-logit TTA instead of no TTAResNet-50 · ImageNet (standard policy)Top-1 accuracy: 75.78 -> 76.91Improved
use mean-logit TTA instead of no TTA with expanded policyCNN-7 · CIFAR100 (expanded policy)Top-1 accuracy: 74.37 -> 73.41Hurt
aggregate with Max instead of Mean under expanded policy (ablation)ResNet-18 · Flowers102 (expanded policy)Top-1 accuracy: 89.47 -> 83.30Hurt
use greedy policy search (GPS) instead of Mean (ablation)MobileNetV2 · Flowers102 (standard policy)Top-1 accuracy: 90.47 -> 88.28Hurt
small model with learned TTA vs larger model without TTAMobileNetV2 vs ResNet-50 · Flowers102 (standard policy)Top-1 accuracy: 91.72 -> 92.62Improved
expanded instead of standard TTA policy for learned aggregation (ablation)learned TTA aggregation, MobileNetV2 · Flowers102Top-1 accuracy: 92.62 -> 92.49No clear effect

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