Reproduce “DeCoOp: Robust Prompt Tuning with Out-of-Distribution Detection” 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
Zhi Zhou, Ming Yang, Jiang-Xin Shi, Lan-Zhe Guo, Yu-Feng Li
DeCoOp addresses open-world prompt tuning, where a vision-language classifier trained on base classes must classify a mixed set of base and new classes. It routes images judged as base to a tuned classifier and others to a zero-shot classifier. Across 11 datasets with ViT-B/16, it achieves 69.69% full-space accuracy versus CoCoOp’s 67.67%, under a few-shot protocol with known new-class names.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.