Research map / Detection and segmentation / CNN features
Rethinking Detection Calibration: A Coordinate Perspective
Juyong Lee, Seungjin Jung, Jungmin Lee, Sunju Lee, Jongwon Choi
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 coordinate-wise confidence re-encoding (ReDC) versus box-level IR calibration | Deformable-DETR · COCO minitest (in-domain) | Cx1-ECE: 12.0 -> 7.7 | Improved |
| ReDC versus box-level calibrators on box-level calibration | Deformable-DETR · COCO minitest (in-domain) | LaECE0: 7.8 -> 7.9 | No clear effect |
| ReDC versus IR for LaECE0 on Cityscapes | Deformable-DETR · Cityscapes minitest (in-domain) | Cx1-ECE: 12.2 -> 6.9 | Improved |
| ReDC under domain shift to COCO-C | Deformable-DETR · COCO-C minitest (domain shift) | Cx1-ECE: 16.1 -> 11.9 | Improved |
| ReDC under domain shift to Foggy Cityscapes | Deformable-DETR · Foggy Cityscapes minitest (domain shift) | Cx1-ECE: 13.7 -> 10.2 | Improved |
| add directional displacement estimation (ablation) | ReDC · COCO (ablation, 5 seeds) | LaECE0: 9.96 -> 9.88 | Improved |
| add directional displacement estimation (ablation) | ReDC · COCO (ablation, 5 seeds) | LaACE0: 23.14 -> 23.14 | No clear effect |
| add directional displacement estimation (ablation) | ReDC · COCO (ablation, 5 seeds) | Cx1-ECE: 11.22 -> 11.18 | No clear effect |
| add directional displacement estimation (ablation) | ReDC · COCO (ablation, 5 seeds) | Cy1-ECE: 11.90 -> 10.86 | Improved |
| ReDC versus probabilistic-detector calibration GP-Normal | ReDC · COCO coordinate-wise error correlation | mean correlation: 0.0263 -> 0.1771 | Improved |
| apply ReDC to a different detector (DINO) | DINO · COCO (model-agnostic) | Cx1-ECE: 18.3 -> 9.4 | Improved |
| apply ReDC to Cascade R-CNN | Cascade R-CNN · COCO (model-agnostic) | Da-CE: 19.3 -> 28.1 | Hurt |
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.
- Cal-DETR — Cal-DETR: Calibrated Detection Transformer (NeurIPS 2023) · 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
- Multiclass Confidence and Localization Calibration for Object Detection (CVPR 2023 · similarity 0.92)
- Bridging Precision and Confidence: A Train-Time Loss for Calibrating Object Detection (CVPR 2023 · similarity 0.91)
- "On Calibration of Object Detectors: Pitfalls, Evaluation and Baselines" (ECCV 2024 · similarity 0.90)
- Systematic Evaluation of Uncertainty Calibration in Pretrained Object Detectors (IJCV 2024 · similarity 0.89)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · similarity 0.86)
- Improving Single Domain-Generalized Object Detection: A Focus on Diversification and Alignment (CVPR 2024 · similarity 0.86)
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.
