New method LE-GDA boosts CLIP adaptation using intermediate representations

Researchers have developed LE-GDA, a novel method for training-free few-shot adaptation of vision-language models like CLIP. This approach leverages intermediate representations from the model's encoder, which are typically unused in standard methods. By fitting regularized discriminants to these representations and combining them, LE-GDA standardizes class scores and improves accuracy across various datasets and CLIP backbones. The method demonstrated a mean accuracy gain of 1.53% at 16 shots and a 1.20-point gain at one shot compared to a baseline that only uses final-stage embeddings, highlighting the value of intermediate representations. AI

IMPACT Enhances few-shot adaptation capabilities for vision-language models by effectively utilizing intermediate representations.

RANK_REASON The cluster contains an academic paper detailing a new method for adapting vision-language models. [lever_c_demoted from research: ic=1 ai=1.0]

Read on arXiv cs.AI →

• Gaussian Discriminant Analysis for Optimal Delineation of Mild Cognitive Impairment in Alzheimer's Disease
• GDA-CLIP
• LE-GDA

• paper
• model release

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →