Research map / Detection and segmentation / Open-set and few-shot detection
RbA: Segmenting Unknown Regions Rejected by All
Nazir Nayal, Misra Yavuz, João F. Henriques, Fatma Güney
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 |
|---|---|---|---|
| Fine-tune RbA with outlier supervision | RbA · SMIYC Anomaly Track | AP: 86.1 -> 90.9 | Improved |
| Use RbA scoring instead of DenseHybrid with outlier supervision | RbA · SMIYC Anomaly Track | FPR@95: AP 78.0 -> 90.9; FPR@95 9.8 -> 11.6 | Mixed |
| Fine-tune with RbA loss instead of using PEBAL on Mask2Former | Mask2Former · Cityscapes closed-set | mIoU: 75.32 -> 82.20 | Improved |
| Use squared hinge loss instead of KL divergence, MSE, L1 or BCE (ablation) | RbA · Road Anomaly | AP: 80.94 -> 85.42 (best alternative L1) | Improved |
| Use MSE or L1 loss instead of squared hinge (ablation) | RbA · Fishyscapes LaF | FPR@95: 6.30 -> 22.06 (MSE) | Hurt |
| Fine-tune with RbA across backbones (ablation) | Mask2Former · Road Anomaly (R101) | AP: 38.1 -> 61.9 | Improved |
| Use one transformer decoder layer instead of nine (ablation) | Mask2Former · Road Anomaly | AP and FPR@95 | Improved |
| Increase outlier exposure probability from 0.1 to 0.4 (ablation) | RbA fine-tuning · Cityscapes mIoU | mIoU: 82.15 -> 81.27 | Hurt |
| Fine-tune for more than 3000 iterations (ablation) | RbA fine-tuning · Road Anomaly | AP: 85.72 -> 84.66 (3000 -> 5000) | Hurt |
| Increase number of object queries (ablation) | RbA · Road Anomaly | AP | Improved |
Closest papers by meaning
- A Likelihood Ratio-Based Approach to Segmenting Unknown Objects (IJCV 2025 · similarity 0.89)
- Feature Purification Matters: Suppressing Outlier Propagation for Training-Free Open-Vocabulary Semantic Segmentation (ICCV 2025 · similarity 0.88)
- Mask2Anomaly: Mask Transformer for Universal Open-Set Segmentation (TPAMI 2024 · similarity 0.86)
- A Dataset for Semantic Segmentation in the Presence of Unknowns (CVPR 2025 · similarity 0.86)
- Segment Every Out-of-Distribution Object (CVPR 2024 · similarity 0.85)
- Generalize or Detect? Towards Robust Semantic Segmentation Under Multiple Distribution Shifts (NeurIPS 2024 · similarity 0.85)
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.
