AI Paper Atlas

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

Learning to Prompt for Vision-Language Models

Kaiyang Zhou, Jingkang Yang, Chen Change Loy, Ziwei Liu

IJCV 2022 · Vision-language and multimodal › CLIP and zero-shot · approach: Prompt learning for CLIP · 3,724 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 hand-crafted prompt with learned context (CoOp, M=16)Zero-Shot CLIP ResNet-50 · ImageNetaccuracy: 58.18 -> 62.95Improved
use fewer context tokens (M=4 vs 16) for domain shift (ablation)CoOp ResNet-50 · ImageNet-Sketchaccuracy: 32.74 -> 34.67Improved
apply CoOp (M=4) to ImageNet-A transferCLIP ResNet-50 · ImageNet-Aaccuracy: 21.65 -> 23.06Improved
replace zero-shot with linear probeCLIP ResNet-50 · ImageNet-Sketchaccuracy: 33.32 -> 19.07Hurt
use learned CoOp prompts instead of prompt ensemblingCLIP ResNet-50 · ImageNetaccuracy: 60.41 -> 62.95Improved
use prompt ensembling instead of single engineered promptZero-Shot CLIP ViT-B/16 · ImageNetaccuracy: 66.73 -> 68.74Improved
fine-tune CLIP image encoder instead of prompt learningCLIP ResNet-50 · ImageNet (16 shots)accuracy: 58.18 -> 18.28Hurt
add learnable transformation layer on text outputZero-Shot CLIP ResNet-50 · ImageNet (16 shots)accuracy: 58.18 -> 58.86Improved
add learnable bias on text outputZero-Shot CLIP ResNet-50 · ImageNet (16 shots)accuracy: 58.18 -> 60.93Improved
manual vs random initialization of context (ablation)CoOp · Average over 11 datasetsaccuracy: 72.65 -> 72.65No clear effect
add CoOp learned context to CLIP ResNet-50 on DOSCO-2kCLIP ResNet-50 · DOSCO-2kaccuracy: 60.9 -> 64.7Improved
use CoCoOp instead of CoOp on DOSCO-2kCoOp ViT-B/16 · DOSCO-2kaccuracy: 73.3 -> 72.7Hurt

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