Research map / Label-efficient and robust learning / OOD and anomaly detection
Mind the Way You Select Negative Texts: Pursuing the Distance Consistency in OOD Detection with VLMs
Zhikang Xu, Qianqian Xu, Zitai Wang, Cong Hua, Sicong Li, Zhiyong Yang, Qingming Huang
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 |
|---|---|---|---|
| select negative texts by inter-modal distance instead of intra-modal distance (ablation) | NegLabel · ImageNet-1k Four-OOD avg | FPR95: 25.40 -> 24.12 | Improved |
| add inter-modal guided extra negative embeddings from inverted OOD images (ablation) | Intra SNT · ImageNet-1k Four-OOD avg | FPR95: 25.40 -> 14.76 | Improved |
| use inter-modal instead of intra-modal filtering of extra negative embeddings (ablation) | InterNeg with Inter SNT · ImageNet-1k Four-OOD avg | FPR95: 19.99 -> 14.04 | Improved |
| use inter-modal selection for negative texts on top of inter-modal extra embeddings (ablation) | InterNeg · ImageNet-1k Four-OOD avg | FPR95: 14.76 -> 14.04 | Improved |
| InterNeg vs best prior zero-shot method AdaNeg | CLIP ViT-B/16 · ImageNet-1k Four-OOD avg | FPR95: 18.92 -> 14.04 | Improved |
| InterNeg vs AdaNeg on harder benchmark | CLIP ViT-B/16 · ImageNet-1k OpenOOD Near-OOD | AUROC: 76.70 -> 82.20 | Improved |
| use only one ID image per class as proxy (ablation) | InterNeg · ImageNet-1k Four-OOD | AUROC/FPR95 | Hurt |
| increase K beyond M=2000 (ablation) | InterNeg · Four-OOD and Near-OOD | OOD detection performance | Hurt |
| vary high-confidence threshold beta (ablation) | InterNeg · Four-OOD and Near-OOD | OOD detection performance | No clear effect |
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
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models (ICLR 2024 · similarity 0.91)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.91)
- AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.90)
- WikiDO: A New Benchmark Evaluating Cross-Modal Retrieval for Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · similarity 0.88)
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection (ICML 2024 · 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.
