AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Black Box Few-Shot Adaptation for Vision-Language Models

Yassine Ouali, Adrian Bulat, Brais Matinez, Georgios Tzimiropoulos

ICCV 2023 · Vision-language and multimodal › CLIP and zero-shot · approach: Test-time adaptation of VLMs · 41 citations (OpenAlex)

Read the paper

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 soft-prompt learning (CoOp) with Linear Feature AlignmentCLIP ViT-B/16 · 11 few-shot datasets avg (16-shot)top-1 accuracy: 79.71 -> 81.21Improved
replace CoOp with LFA on training costImageNet ViT-B/16 · ImageNet 16-shottraining time: 3h 22min -> 3min 9sImproved
add beta regularization to orthogonal Procrustes (ablation)LFA · ImageNet 16-shottop-1 accuracy: 52.5 -> 64.8Improved
add ARerank refinement on top of beta-Procrustes (ablation)LFA · ImageNet 16-shottop-1 accuracy: 64.8 -> 71.7Improved
use ARerank loss instead of CSLS (ablation)LFA refinement · ImageNet 16-shottop-1 accuracy: 71.1 -> 71.7Improved
use ARerank loss instead of contrastive loss (ablation)LFA refinement · Aircraft 16-shottop-1 accuracy: 40.1 -> 45.1Improved
add beta-Procrustes initialization before refinement (ablation)LFA · ImageNet 16-shottop-1 accuracy: 71.6 -> 71.7No clear effect
vary beta between 0.6 and 0.9 (ablation)LFA · ImageNet 16-shottop-1 accuracy: 72.45 -> 72.54No clear effect
use 5 crops and prompt template (ablation)LFA · 11 datasets avg (ViT-B/16)top-1 accuracy: 80.54 -> 81.21Improved
combine LFA with CoOp soft-prompt prototypes (ablation)LFA · ImageNet 16-shot (ViT-B/16)top-1 accuracy: 72.61 -> 73.10Improved
use two mappings (W and EMA mapping) for base-to-new generalization (ablation)LFA · ImageNet new classestop-1 accuracy: 67.1 -> 69.4Improved
use LFA for unsupervised adaptation (U-LFA, n=5) without labelsCLIP ViT-B/16 · ImageNettop-1 accuracy: 62.8 -> 68.6Improved

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