AI Paper Atlas

Research map / Vision-language and multimodal / Cross-modal learning

Multimodality Helps Unimodality: Cross-Modal Few-Shot Learning With Multimodal Models

Zhiqiu Lin, Samuel Yu, Zhiyi Kuang, Deepak Pathak, Deva Ramanan

CVPR 2023 · Vision-language and multimodal › Cross-modal learning · approach: Unimodal-to-multimodal representation transfer · 126 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
Add class-name text samples to image few-shot trainingCLIP linear probing · 11-dataset average, 1-shottop-1 accuracy: 36.7 -> 64.1Improved
Add class-name text samples and partially finetune the image encoderCLIP few-shot adaptation · 11-dataset average, 1-shottop-1 accuracy: 59.6 -> 64.7Improved
Add audio training samples to image classificationImageNet-ESC-19 image classifier · ImageNet-ESC-19, 1-shotaccuracy: 68.0 -> 69.3Improved
Add audio training samples to image classificationImageNet-ESC-27 image classifier · ImageNet-ESC-27, 4-shotaccuracy: 79.0 -> 78.9Hurt
Add image training samples to audio classificationImageNet-ESC-19 audio classifier · ImageNet-ESC-19, 1-shotaccuracy: 31.2 -> 35.7Improved
Add image training samples to audio classificationImageNet-ESC-27 audio classifier · ImageNet-ESC-27, 1-shotaccuracy: 28.2 -> 35.0Improved
Use cross-modal training rather than image-only trainingWiSE-FT · 11-dataset average, 1-shottop-1 accuracy: 59.1 -> 63.8Improved
Use cross-modal training rather than image-only trainingCoOp prompting · 11-dataset average, 1-shottop-1 accuracy: 59.6 -> 62.0Improved
Use cross-modal training rather than image-only trainingTip-Adapter · 11-dataset average, 1-shottop-1 accuracy: 63.3 -> 64.4Improved
Use text-feature initialization instead of random initialization (ablation)Cross-Modal Linear Probing · 11-dataset average, 1-shottop-1 accuracy: 48.37_{(1.58)} -> 63.66_{(1.25)}Improved
Use text-feature initialization instead of random initialization (ablation)Cross-Modal Partial Finetuning · 11-dataset average, 1-shottop-1 accuracy: 42.03_{(1.91)} -> 64.27_{(0.96)}Improved
Finetune image attention pooling while freezing text encoder (ablation)Cross-modal partial finetuning · 11-dataset average, 16-shottop-1 accuracy: 75.54_{(0.53)} -> 76.65_{(0.44)}Improved

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.

Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.

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