Reproduce “Guided Prompt Evolution for Vision-Language Models Adaptation” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Enming Zhang, Jiayang Li, Yanlong Wang, Yanru Wu, Zhenyu Liu, Yang Li
EvoPrompt adapts CLIP prompts for few-shot image classification by generating layer- and modality-specific prompts from shared structure and constraining low-rank updates. It freezes update directions while adjusting their magnitudes, and uses a constancy loss. Across 11 datasets, 16 examples per base class yield averages of 85.90% Base, 78.35% Novel, and 81.95% HM; this does not establish preservation of all zero-shot capabilities.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.