AI Paper Atlas

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

Multi-Label Test-Time Adaptation with Bayesian Conditional Priors

QiRu Li, Ao Zhou, Zhiwei Jiang, Zifeng Cheng, Cong Wang, Yafeng Yin, Qing Gu

ICML 2026 · Label-efficient and robust learning › Test-time adaptation · approach: Test-time adaptation of vision-language models · 0 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
use 2 anchors instead of 1 (average PMI over anchors) (ablation)BCP ViT-B/16 · Average of 5 datasetsmAP: 71.79 -> 69.93Hurt
use 3 anchors instead of 1 (average PMI over anchors) (ablation)BCP ViT-B/16 · Average of 5 datasetsmAP: 71.79 -> 67.94Hurt
add warmup that delays correction until statistics stabilize (ablation)BCP ViT-B/16 · Average of 5 datasetsmAP: 71.79 -> 71.95Improved
apply BCP prior correction to frozen CLIP on Anchor-Hit samplesCLIP ViT-B/16 · Anchor-Hit subset averagemAP: 72.73 -> 80.33Improved
apply BCP prior correction to frozen CLIP on Anchor-Miss samplesCLIP ViT-B/16 · Anchor-Miss subsetmAP: +1.48 on COCO2014, -3.75 on VOC2007Mixed
use BCP instead of BEMCLIP RN50 · COCO2014adaptation time and mAP: 0.24s -> 0.112s; 51.58 -> 61.69Improved
use BCP instead of SCACLIP ViT-B/16 · Average of 5 datasetsmAP: 69.24 -> 71.79Improved
use BCP instead of DOTACLIP RN50 · NUSWIDEmAP: 38.88 -> 50.34Improved
use BCP instead of SCA on tail labelsCLIP ViT-B/16 · COCO14 tail splitmean AP: 67.61 -> 69.22Improved
apply BCP to CLIP with many labels per imageCLIP RN50 · COCO2014 labels >= 8mAP: 41.07 -> 51.64Improved
make test stream more non-i.i.d. (gamma 0.1 to 0.001) (ablation)BCP ViT-B/16 · Average of 5 datasetsmAP: 69.15 -> 68.66No clear effect
shuffle test stream order with five seeds (ablation)BCP ViT-B/16 · COCO14mAP: 65.48, 65.40, 65.48, 65.49, 65.42No 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