Reproduce “CLIPN for Zero-Shot OOD Detection: Teaching CLIP to Say No” 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
Hualiang Wang, Yi Li, Huifeng Yao, Xiaomeng Li
The paper presents CLIPN, a zero-shot out-of-distribution detector that adds learned “no” text representations to CLIP, comparing images with both known-class descriptions and their negations. It is pretrained on paired web image-text data. Its CLIPN-A variant reaches 93.10 AUROC and 31.10 FPR95 on ImageNet-1K, averaged across four OOD datasets, versus MCM’s 90.76 and 42.74. These results concern image-classification settings.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.