Research map / Detection and segmentation / Open-set and few-shot detection
Unmasking Anomalies in Road-Scene Segmentation
Shyam Nandan Rai, Fabio Cermelli, Dario Fontanel, Carlo Masone, Barbara Caputo
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 global masked attention to Mask2Former (ablation) | Mask2Anomaly · FS Lost & Found val | AuPRC: 10.60 -> 32.35 | Improved |
| use masked attention instead of cross-attention (ablation) | Mask2Former · FS Lost & Found val | AuPRC: 20.30 -> 10.60 | Hurt |
| add mask contrastive learning with outlier masks (ablation) | Mask2Anomaly · FS Lost & Found val | FPR95: 87.11 -> 31.93 (with RM, with/without GMA) | Improved |
| add refinement mask filtering stuff regions except road (ablation) | Mask2Anomaly · FS Lost & Found val | FPR95: 24.07 -> 9.46 | Improved |
| filter things instead of stuff in refinement mask (ablation) | Mask2Anomaly · FS Lost & Found val | FPR95: 24.07 -> 39.11 | Hurt |
| set contrastive margin to 0.75 rather than 0.5 or 1 (ablation) | Mask2Anomaly · FS Lost & Found val | AuPRC: 62.07 (m=0.5) -> 69.41 | Improved |
| raise batch outlier probability from 0.1 to 0.2 (ablation) | Mask2Anomaly · FS Lost & Found val | AuPRC: 63.01 -> 69.41 | Improved |
| cast anomaly segmentation as mask classification via Mask2Anomaly vs per-pixel Dense Hybrid | Mask2Anomaly · Average over five benchmarks | FPR95: 16.79 -> 6.68 | Improved |
| Mask2Anomaly full model vs vanilla Mask2Former on in-distribution segmentation | Mask2Anomaly · Cityscapes val | mIoU: 78.88 vs 1.46 points above Mask2Former | Improved |
| use larger backbones for Mask2Former instead of Mask2Anomaly ResNet-50 (ablation) | Mask2Former · FS Lost & Found val | AuPRC: ResNet-50 10.60 -> Swin-S 30.96 vs Mask2Anomaly 32.35 | No clear effect |
| Mask2Anomaly on Road Anomaly versus JSRNet | Mask2Anomaly · Road Anomaly | AuPRC: 94.4 (JSRNet) vs 79.70 | Hurt |
Closest papers by meaning
- Mask2Anomaly: Mask Transformer for Universal Open-Set Segmentation (TPAMI 2024 · similarity 0.94)
- Pixel-Wise Anomaly Detection in Complex Driving Scenes (CVPR 2021 · similarity 0.90)
- Spotting the Unexpected (STU): A 3D LiDAR Dataset for Anomaly Segmentation in Autonomous Driving (CVPR 2025 · similarity 0.87)
- Self-Supervised Likelihood Estimation with Energy Guidance for Anomaly Segmentation in Urban Scenes (AAAI 2024 · similarity 0.87)
- A Dataset for Semantic Segmentation in the Presence of Unknowns (CVPR 2025 · similarity 0.86)
- AnomalyNCD: Towards Novel Anomaly Class Discovery in Industrial Scenarios (CVPR 2025 · similarity 0.84)
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.
