Research map / Label-efficient and robust learning / OOD and anomaly detection
Can OOD Object Detectors Learn from Foundation Models?
Jiahui Liu, Xin Wen, Shizhen Zhao, Yingxian Chen, Xiaojuan Qi
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 similarity-based filter for selecting synthetic OOD objects (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 39.29 -> 36.44 | Improved |
| add SAM-based box refiner (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 39.55 -> 36.44 | Improved |
| inpaint ID objects in scene images instead of generating object-centric images (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 51.99 -> 36.44 | Improved |
| use object boxes rather than whole edited scene image as OOD sample (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 48.01 -> 36.44 | Improved |
| increase number of synthetic samples from 4k to 14k (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 36.44 -> 36.70 | No clear effect |
| vary number of imagined novel concepts per ID object (3 to 8) (ablation) | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | FPR95: 36.96 to 37.91 | No clear effect |
| train SyncOOD OOD detector on top of Faster R-CNN instead of SAFE | Faster R-CNN OOD detection · PASCAL-VOC (ID) vs MS-COCO | FPR95: 50.86 -> 36.44 | Improved |
| SyncOOD vs DFDD on AUROC for PASCAL-VOC / MS-COCO | SyncOOD · PASCAL-VOC (ID) vs MS-COCO | AUROC: 90.79 vs 86.52 | Hurt |
Closest papers by meaning
- Image Background Serves as Good Proxy for Out-of-distribution Data (ICLR 2024 · similarity 0.91)
- RUNA: Object-Level Out-of-Distribution Detection via Regional Uncertainty Alignment of Multimodal Representations (AAAI 2025 · similarity 0.89)
- COUNTS: Benchmarking Object Detectors and Multimodal Large Language Models under Distribution Shifts (CVPR 2025 · similarity 0.89)
- Is Complex Training Necessary for Long-Tailed OOD Detection? A Re-think from Feature Geometry (NeurIPS 2026 · similarity 0.89)
- Conjugated Semantic Pool Improves OOD Detection with Pre-trained Vision-Language Models (NeurIPS 2024 · similarity 0.88)
- ImageNet-OOD: Deciphering Modern Out-of-Distribution Detection Algorithms (ICLR 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.
