AI RESEARCH
Enhancing Few-Shot Out-of-Distribution Detection via the Refinement of Foreground and Background
arXiv CS.CV
•
ArXi:2601.15065v2 Announce Type: replace CLIP-based foreground-background (FG-BG) decomposition methods have nstrated remarkable effectiveness in improving few-shot out-of-distribution (OOD) detection performance. However, existing approaches still suffer from several limitations. For background regions obtained from decomposition, existing methods adopt a uniform suppression strategy for all patches, overlooking the varying contributions of different patches to the prediction.