AI Paper Atlas

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

Regularized Mask Tuning: Uncovering Hidden Knowledge in Pre-Trained Vision-Language Models

Kecheng Zheng, Wei Wu, Ruili Feng, Kai Zhu, Jiawei Liu, Deli Zhao, Zheng-Jun Zha, Wei Chen, Yujun Shen

ICCV 2023 · Vision-language and multimodal › CLIP and zero-shot · approach: Parameter-efficient visual tuning · 11 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
Learn binary masks on the attention layers of frozen CLIP (R-AMT) instead of zero-shot useCLIP ViT-B/16 · 11 datasets, 16-shot averageaccuracy (%): 65.23 -> 83.96Improved
Add R-AMT mask to CoOPCoOP · 11 datasets, 16-shot averageaccuracy (%): 79.90 -> 83.16Improved
Add R-AMT mask to TIP-AdapterTIP-Adapter · 11 datasets, 16-shot averageaccuracy (%): 81.24 -> 84.37Improved
Add gradient dropout regularity to mask tuning (ablation)AMT · Base-to-new, new classesaccuracy (%): 69.11 -> 72.15Improved
Add KL loss directly to the CE loss instead of gradient dropout (ablation)AMT · ImageNet 16-shotaccuracy (%): 72.60 -> 71.92Hurt
Use AgreeGrad gradient surgery with AMT (ablation)AMT · ImageNet 16-shotaccuracy (%): 72.60 -> 68.82Hurt
Use ProGrad gradient projection with AMT (ablation)AMT · ImageNet 16-shotaccuracy (%): 72.60 -> 72.70Improved
Lower leak parameter l from 1.0 to 0.1 (ablation)R-AMT · ImageNet 16-shotaccuracy (%): 73.07 -> 72.67Hurt
Mask only attention layers instead of all layers (ablation)R-PMT · 11 datasets, 16-shot averageaccuracy (%): 83.96 vs 83.96, storage 19M -> 6.7MNo clear effect
Mask attention layers instead of MLP layers (ablation)R-MMT · 11 datasets, 16-shot averageaccuracy (%): 83.39 -> 83.96Improved
Set hard threshold alpha to 5e-3 rather than 4e-3 or 6e-3 (ablation)R-AMT · ImageNet 16-shotaccuracy (%): 72.87 / 72.91 -> 73.07Improved
Fine-tune whole image encoder instead of mask tuningCLIP image encoder · ImageNet 16-shotaccuracy (%): 66.73 (zero-shot) -> 64.51Hurt

Earlier methods this paper compares with

Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 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