Research map / Vision-language and multimodal / CLIP and zero-shot
How Does Fine-Tuning Impact Out-of-Distribution Detection for Vision-Language Models?
Yifei Ming, 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 |
|---|---|---|---|
| fine-tune CLIP with CoCoOp prompt learning (ablation) | ZOCLIP · Caltech-101 (ID) | avg FPR95 with S_MS: 37.96 -> 5.94 | Improved |
| fine-tune CLIP with CoCoOp (ablation) | ZOCLIP · Caltech-101 (ID) | avg FPR95 with S_MCM: 15.28 -> 5.02 | Improved |
| fine-tune CLIP with CoOp (ablation) | ZOCLIP · Food-101 (ID) | avg FPR95 with S_MS: 1.24 -> 0.36 | Improved |
| fine-tune CLIP with CoOp (ablation) | ZOCLIP · Stanford-Cars (ID) | avg FPR95 with S_MS: 0.07 -> 0.02 | No clear effect |
| use S_MCM instead of S_MS (softmax scaling with temperature) (ablation) | TipAdaptor · Caltech-101 (ID) | avg FPR95: 13.86 -> 6.19 | Improved |
| use S_MCM instead of S_MS (ablation) | CoCoOp · Caltech-101 (ID) | avg FPR95: 5.94 -> 5.02 | Improved |
| use S_MS instead of default-temperature S_MSP (ablation) | CoOp · Caltech-101 (ID) | avg FPR95: 26.66 -> 8.35 | Improved |
| use S_MCM instead of S_MSP (ablation) | CoOp, CLIP-B/32 · Caltech-101 (ID) | avg FPR95: 26.40 -> 6.17 | Improved |
| use S_MCM instead of S_MSP (ablation) | CoOp · ImageNet-O (ID ImageNet-1k) | FPR95: 77.20 -> 61.50 | Improved |
| use S_MCM instead of S_MS (ablation) | CoOp · ImageNet-1k (ID) | avg FPR95: 7.44% lower | Improved |
| scale backbone from RN50 to CLIP-L/14 (ablation) | CoOp with S_MCM · Caltech-101 (ID) | avg FPR95: 14.67 -> 3.50 | Improved |
| use S_MS instead of S_MSP on ViT versus ResNet (ablation) | CoOp · Caltech-101 (ID) | avg FPR95: RN101 27.11 -> 26.93; CLIP-B/32 26.40 -> 17.13 | Mixed |
Closest papers by meaning
- CRoFT: Robust Fine-Tuning with Concurrent Optimization for OOD Generalization and Open-Set OOD Detection (ICML 2024 · similarity 0.92)
- Overcoming the Pitfalls of Vision-Language Model Finetuning for OOD Generalization (ICLR 2024 · similarity 0.92)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.92)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Weak Distribution Detectors Lead to Stronger Generalizability of Vision-Language Prompt Tuning (AAAI 2024 · similarity 0.90)
- Self-Calibrated Tuning of Vision-Language Models for Out-of-Distribution Detection (NeurIPS 2024 · similarity 0.90)
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.
