Research map / Label-efficient and robust learning / OOD and anomaly detection
A Novel Fine-Tuned CLIP-OOD Detection Method with Double Loss Constraint Through Optimal Transport Semantic Alignment
Hengyang Lu, Xin Guo, Shuai Feng, Wenyu Jiang, Yuntao Du, Chang Xia, Chenyou Fan
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 Lalign to Lce baseline (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 18.55 -> 18.49 | No clear effect |
| add Lalign to Lce baseline (ablation) | DOT-OOD · ImageNet-1k ID accuracy | ID accuracy: 71.99 -> 72.40 | Improved |
| add Lneg to Lce baseline (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 18.55 -> 15.90 | Improved |
| add both Lalign and Lneg (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 18.55 -> 14.09 | Improved |
| add both Lalign and Lneg (ablation) | DOT-OOD · ImageNet-1k ID accuracy | ID accuracy: 71.99 -> 74.04 | Improved |
| add OT module (vs replacing enhanced features with original) (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 23.66 -> 14.09 | Improved |
| add OT module (ablation) | DOT-OOD · ImageNet-1k ID accuracy | ID accuracy: 69.39 -> 74.04 | Improved |
| without OT, losses on original features vs Lce baseline (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 18.55 -> 23.66 | Hurt |
| fine-tune deep layers instead of only projection layers (ablation) | DOT-OOD · ImageNet-1k avg OOD | FPR95: 16.98 -> 14.09 | Improved |
| fine-tune deep layers instead of middle layers (ablation) | DOT-OOD · ImageNet-1k ID accuracy | ID accuracy: 73.66 -> 74.04 | Improved |
| integrate DOT-OOD fine-tuning into MCM score | MCM, CLIP-B/16 · ImageNet-1k avg OOD | FPR95: 42.61 -> 32.93 | Improved |
| integrate DOT-OOD fine-tuning into NegLabel score | NegLabel, CLIP-B/16 · ImageNet-1k avg OOD | FPR95: 25.40 -> 16.06 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- Local-Prompt — Local-Prompt: Extensible Local Prompts for Few-Shot Out-of-Distribution Detection (ICLR 2025) · better on 5 of 6 shared results
- LoCoOp — LoCoOp: Few-Shot Out-of-Distribution Detection via Prompt Learning (NeurIPS 2023) · better on 2 of 2 shared results
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.87)
- RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025 · similarity 0.87)
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 2024 · similarity 0.87)
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · similarity 0.87)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.86)
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.
