Research map / Vision-language and multimodal / CLIP and zero-shot
Delving into Out-of-Distribution Detection with Vision-Language Representations
Yifei Ming, Ziyang Cai, Jiuxiang Gu, Yiyou Sun, Wei Li, Yixuan Li
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add softmax scaling (tau=1) to max cosine similarity score (ablation) | MCM · ImageNet-1k (ID) | FPR95: 69.08 -> 42.74 | Improved |
| add softmax scaling to max cosine similarity score (ablation) | MCM · ImageNet-20 (ID) | FPR95: 13.93 -> 2.60 | Improved |
| add softmax scaling to max cosine similarity score (ablation) | MCM · Food-101 / Stanford-Cars / Oxford-Pet | FPR95: 2.37 -> 1.86 (Food-101) | No clear effect |
| use text-embedding prototypes (MCM) instead of visual-only Mahalanobis prototypes | CLIP-B image encoder · ImageNet-1k (ID) | AUROC: 73.14% -> 90.77% | Improved |
| scale CLIP-B to CLIP-L (ablation) | MCM · ImageNet-1k (ID) | FPR95: 42.74 -> 38.17 | Improved |
| replace ViT image encoder with ResNet (RN50x4) (ablation) | MCM · ImageNet-1k (ID) | AUROC: 90.77 -> 89.97 | No clear effect |
| use ensemble of five prompts instead of default prompt (ablation) | MCM (CLIP-L) · ImageNet-1k (ID) | FPR95: 38.17% -> 35.23% | Improved |
| fine-tune with linear probing and use MSP instead of zero-shot MCM | CLIP-L · Waterbirds spurious OOD | FPR95: 5.87 -> 39.57 | Hurt |
| use MCM zero-shot instead of Mahalanobis | CLIP-B/16 · ImageNet-10 vs ImageNet-20 | FPR95: 78.32 -> 5.00 | Improved |
Closest papers by meaning
- Out-of-Distribution Detection with Positive and Negative Prompt Supervision Using Large Language Models (AAAI 2026 · similarity 0.92)
- How Does Fine-Tuning Impact Out-of-Distribution Detection for Vision-Language Models? (IJCV 2023 · similarity 0.92)
- Intra-Image Mining and Symmetric Maximum Concept Matching for Few Shot Out-of-Distribution Detection (AAAI 2026 · similarity 0.92)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.91)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.89)
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.
