Research map / Label-efficient and robust learning / OOD and anomaly detection
Synthesizing Near-Boundary OOD Samples for Out-of-Distribution Detection
Jinglun Li, Kaixun Jiang, Zhaoyu Chen, Bo Lin, Yao Tang, Weifeng Ge, Wenqiang 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 iterative gradient-guided generation instead of direct negative-label prompts (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 94.85 -> 96.21 | Improved |
| use iterative gradient-guided generation instead of direct negative-label prompts (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | FPR95: 22.59 -> 17.29 | Improved |
| add negative label feature fine-tuning to gradient-generated data (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 96.21 -> 97.01 | Improved |
| add negative label feature fine-tuning to gradient-generated data (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | FPR95: 17.29 -> 14.27 | Improved |
| add negative label fine-tuning when using direct negative-label images (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 94.85 -> 95.01 | Improved |
| add projection layer trained on direct negative-label images (ablation) | NegLabel · ImageNet-1k avg 4 OOD sets | AUROC: 94.21 -> 94.85 | Improved |
| add projection layer trained on gradient-generated images (ablation) | NegLabel · ImageNet-1k avg 4 OOD sets | FPR95: 25.40 -> 17.29 | Improved |
| train projection and label features separately instead of jointly (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | FPR95: 19.41 -> 14.27 | Improved |
| exclude the trained projection layer during label feature fine-tuning (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 96.49 -> 97.01 | Improved |
| increase synthetic sample count m beyond 50k (ablation) | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC | Hurt |
| apply SynOOD on CLIP ResNet50 versus NegLabel | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 92.97 -> 95.80 | Improved |
| apply SynOOD on CLIP ViT-B/32 versus NegLabel | SynOOD · ImageNet-1k avg 4 OOD sets | AUROC: 93.67 -> 95.93 | 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 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
- GOOD: Training-Free Guided Diffusion Sampling for Out-of-Distribution Detection (NeurIPS 2025 · similarity 0.91)
- Enhanced OoD Detection through Cross-Modal Alignment of Multi-Modal Representations (CVPR 2025 · similarity 0.90)
- Overcoming Shortcut Problem in VLM for Robust Out-of-Distribution Detection (CVPR 2025 · similarity 0.89)
- Envisioning Outlier Exposure by Large Language Models for Out-of-Distribution Detection (ICML 2024 · similarity 0.89)
- MultiOOD: Scaling Out-of-Distribution Detection for Multiple Modalities (NeurIPS 2024 · similarity 0.89)
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 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.
