Research map / Vision-language and multimodal / CLIP and zero-shot
Test-Time Adaptation via Self-Reinforced Optimal Transport for Zero-Shot OOD Detection with Vision–Language Models
MingCai Chen, Heng-yang Lu, Yuntao Du, Baoming Zhang, Hao Zhou
Read the paperCheck your idea against 137,217 papers
AI-extracted summary. Written by AI from the paper's abstract only. Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Proposes a test-time adaptation method using self-reinforced optimal transport to improve zero-shot OOD detection with vision-language models, addressing the unreliable confidence of zero-shot VLM posteriors.
Problem
Zero-shot VLM confidence scores (maximum posterior) are unreliable for out-of-distribution detection, giving much weaker ID/OOD separation than supervised classifiers.
Approach
Uses test-time adaptation driven by a self-reinforced optimal transport formulation to refine VLM posteriors on unlabeled test data, aiming to close the confidence gap with supervised classifiers. Based on the abstract only; the precise mechanism is not described in the available text.
Closest papers by meaning
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Noisy Test-Time Adaptation in Vision-Language Models (ICLR 2025 · similarity 0.89)
- AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- Can We Ignore Labels in Out of Distribution Detection? (ICLR 2025 · similarity 0.88)
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models (ICLR 2024 · similarity 0.88)
- Reflexive Guidance: Improving OoDD in Vision-Language Models via Self-Guided Image-Adaptive Concept Generation (ICLR 2025 · similarity 0.88)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
