Research map / Label-efficient and robust learning / OOD and anomaly detection
A Close Look at Negative Label Guided Out-of-distribution Detection in Pre-trained Vision-Language Models
Bo Peng, Jie Lu, Zhen Fang, Guangquan 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 |
|---|---|---|---|
| use Renyi divergence instead of KL divergence for the worst-case distribution set | NegLabel energy score · ImageNet-1K (CLIP ViT-B/16) | FPR95: 28.13 -> 23.17 | Improved |
| use Renyi divergence instead of KL divergence | NegLabel energy score · ImageNet-1K (CLIP ViT-B/16) | AUROC: 93.77 -> 94.51 | Improved |
| define energy as margin to negative labels instead of raw zero-shot logits | Energy score · ImageNet-1K (CLIP ViT-B/16) | FPR95: 82.21 -> 28.82 | Improved |
| drop in-distribution term from NegLabel denominator (ablation) | NegLabel · ImageNet-1K (CLIP ViT-B/16) | AUROC: 93.76 -> 93.77 | No clear effect |
| KL-based worst-case score versus plain margin energy on Textures | NegLabel · Textures OOD | AUROC: 91.02 -> 89.85 | Hurt |
| use Renyi divergence instead of KL with ViT-B/32 encoder | NegLabel · ImageNet-1K (CLIP ViT-B/32) | FPR95: 27.92 -> 24.79 | Improved |
| use Renyi divergence instead of KL with ResNet50 encoder | NegLabel · ImageNet-1K (CLIP ResNet50) | FPR95: 30.70 -> 27.60 | Improved |
| use Renyi divergence instead of KL with 336x336 input | NegLabel · ImageNet-1K (CLIP ViT-L/14, 336x336) | FPR95: 23.80 -> 21.66 | Improved |
| use Renyi divergence with Common-20K corpus | NegLabel · ImageNet-1K (Common-20K corpus) | FPR95: 48.06 -> 44.64 | Improved |
| use Renyi divergence with Part-of-Speech corpus | NegLabel · ImageNet-1K (Part-of-Speech corpus) | FPR95: 30.59 -> 26.77 | Improved |
| use Renyi divergence on ImageNet-S as ID | NegLabel · ImageNet-S ID (CLIP ViT-B/16) | FPR95: 27.42 -> 21.50 | Improved |
| use Renyi divergence on ImageNet-R as ID | NegLabel · ImageNet-R ID (CLIP ViT-B/16) | FPR95: 20.63 -> 18.30 | 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.
- LAPT — LAPT: Label-driven Automated Prompt Tuning for OOD Detection with Vision-Language Models (ECCV 2024) · better on 2 of 2 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
- Negative Label Guided OOD Detection with Pretrained Vision-Language Models (ICLR 2024 · similarity 0.93)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.91)
- Can We Ignore Labels in Out of Distribution Detection? (ICLR 2025 · similarity 0.90)
- AdaNeg: Adaptive Negative Proxy Guided OOD Detection with Vision-Language Models (NeurIPS 2024 · similarity 0.89)
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection (ICML 2024 · similarity 0.89)
- A Noisy Elephant in the Room: Is Your Out-of-Distribution Detector Robust to Label Noise? (CVPR 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.
