Research map / Label-efficient and robust learning / OOD and anomaly detection
ANTS: Adaptive Negative Textual Space Shaping for OOD Detection via Test-Time MLLM Understanding and Reasoning
Wenjie Zhu, Yabin Zhang, Xin Jin, Wenjun Zeng, Lei Zhang
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 negative image mining before ENS generation (ablation) | ANTS · OpenOOD far-OOD | FPR95: 43.87 -> 19.22 | Improved |
| add negative image mining before ENS generation (ablation) | ANTS · OpenOOD near-OOD | FPR95: 74.48 -> 73.70 | No clear effect |
| use ENS with mined negative images vs NegLabel (ablation) | ANTS · OpenOOD far-OOD | FPR95: 27.34 -> 19.22 | Improved |
| add visually similar ID-class mining before VSNL (ablation) | ANTS · OpenOOD near-OOD | FPR95: 74.36 -> 63.11 | Improved |
| add visually similar ID-class mining before VSNL (ablation) | ANTS · OpenOOD far-OOD | FPR95: 53.82 -> 23.44 | Improved |
| combine ENS and VSNL with fixed lambda 0.5 (ablation) | ANTS · OpenOOD near-OOD | FPR95: 73.70 -> 62.05 | Improved |
| use adaptive lambda instead of 0.5 (ablation) | ANTS · OpenOOD far-OOD | FPR95: 21.65 -> 15.38 | Improved |
| use adaptive lambda instead of 0.5 (ablation) | ANTS · OpenOOD near-OOD | FPR95: 62.05 -> 60.98 | Improved |
| generate visually similar labels for all ID classes (delta=1) (ablation) | ANTS · ImageNet-1k near-OOD | FPR95 | Hurt |
| lengthen generated negative sentences excessively (ablation) | ANTS · ImageNet-1k | FPR95: best at average length 8.4 | Mixed |
| use inexpressive or dissimilar MLLM prompts (ablation) | ANTS · ImageNet-1k | FPR95 | Hurt |
| ANTS vs strongest zero-shot baseline CSP | ANTS · ImageNet-1k avg of 4 OOD sets | FPR95: 17.51 -> 11.20 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdaNeg — AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models (NeurIPS 2024) · better on 6 of 6 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
- AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models (ICLR 2024 · similarity 0.90)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Unsupervised Layer-Wise Score Aggregation for Textual OOD Detection (AAAI 2024 · similarity 0.88)
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 2024 · similarity 0.88)
- Rethinking the Evaluation of Out-of-Distribution Detection: A Sorites Paradox (NeurIPS 2024 · similarity 0.87)
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.
