AI Paper Atlas

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

Test-Time Distribution Normalization for Contrastively Learned Visual-language Models

Yifei Zhou, Juntao Ren, Fengyu Li, Ramin Zabih, Ser-Nam Lim

NeurIPS 2023 · Label-efficient and robust learning › Test-time adaptation · approach: Test-time adaptation of vision-language models · 2 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
add distribution normalization (DN)TCL · ImageNet1K zero-shotAcc@1: 20.0 -> 30.5Improved
add DNCLIP · MSCOCO image-to-textR@1: 52.4 -> 51.7No clear effect
add DNCLIP · MSCOCO text-to-imageR@1: 30.2 -> 33.4Improved
add DN on top of test-time augmentation (TTA)CLIP + TTA · cross-modal retrievalaverage recall: 1.3% average improvementImproved
average DN with vanilla dot product (DN*) instead of DNCLIP + TTA + DN · retrievalstd of improvement: 1.3% -> 0.8%Improved
add DN to CLIPCLIP · Flickr8k-ExpertKendall tau_c: 51.4 -> 54.3Improved
add DN to TCLTCL · Flickr8k-ExpertKendall tau_c: 31.0 -> 42.0Improved
add DN to ALBEFALBEF · Flickr8k-ExpertKendall tau_c: 24.9 -> 34.8Improved
add DN to ALBEFALBEF · ImageNet1K zero-shotAcc@1: 37.3 -> 39.9Improved
add DN to ALBEFALBEF · Pascal-50Smean accuracy: 76.2 -> 75.6Hurt
add DN to TCLTCL · Pascal-50Smean accuracy: 65.6 -> 70.3Improved
estimate mean from 1 training sample instead of 100 (DN*) (ablation)CLIP + DN* · ImageNet1K zero-shotAcc@1: 61.7 -> 60.6Hurt

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