Research map / Detection and segmentation / Open-set and few-shot detection
Towards Building Self-Aware Object Detectors via Reliable Uncertainty Quantification and Calibration
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Aggregate the most certain detections rather than sum all detection uncertainties (ablation) | Object detectors using 1 - detection confidence · SAOD-Gen image-level OOD detection | AUROC: 20.9 -> 94.4 | Helped |
| Use the three lowest detection uncertainties instead of mean uncertainty (ablation) | Object detectors · SAOD-Gen image-level OOD detection | AUROC: 85.8 -> 94.8 | Helped |
| Use 1 - detection confidence rather than Dempster-Shafer uncertainty (ablation) | RS-RCNN · SAOD-Gen image-level OOD detection | AUROC: 30.0 -> 94.8 | Helped |
| Use classification entropy rather than localization uncertainty (ablation) | NLL-RCNN · SAOD-Gen image-level OOD detection | AUROC: 87.7 -> 92.4 | Helped |
| Use pseudo-OOD images from masked validation objects to set the rejection threshold (ablation) | F-RCNN image-level rejection · SAOD-Gen | balanced accuracy: 83.2 -> 87.7 | Helped |
| Use pseudo-OOD images from masked validation objects to set the rejection threshold (ablation) | F-RCNN image-level rejection · SAOD-AV | balanced accuracy: 80.9 -> 91.0 | Helped |
| Add classification entropy to localization entropy, with balanced combination (ablation) | NLL-RCNN · SAOD-Gen image-level OOD detection | second reported performance score: 87.72/81.15 -> 92.20/84.78 | Helped |
| Add classification entropy to localization entropy, with balanced combination (ablation) | ES-RCNN · SAOD-Gen image-level OOD detection | second reported performance score: 86.33/73.61 -> 91.41/81.70 | Helped |
| Use more severe image corruption (ablation) | Object detectors · Gen-OD test | AP: 27.0 -> 20.3 -> 12.8 -> 6.9 | Hurt |
| Use top-k survival rather than confidence-thresholded detections (ablation) | Object detectors · COCO val | LaECE | Hurt |
Closest papers by meaning
- Towards Improving Calibration in Object Detection Under Domain Shift (NeurIPS 2022 · similarity 0.87)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 2024 · similarity 0.86)
- Cocoon: Robust Multi-Modal Perception with Uncertainty-Aware Sensor Fusion (ICLR 2025 · similarity 0.86)
- An Empirical Study of the Generalization Ability of Lidar 3D Object Detectors to Unseen Domains (CVPR 2024 · similarity 0.86)
- OpenAD: Open-World Autonomous Driving Benchmark for 3D Object Detection (NeurIPS 2025 · similarity 0.85)
- SparseAlign: a Fully Sparse Framework for Cooperative Object Detection (CVPR 2025 · 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.