Research map / Detection and segmentation / CNN features
"On Calibration of Object Detectors: Pitfalls, Evaluation and Baselines"
Selim Kuzucu, Kemal Oksuz, Jonathan Sadeghi, Puneet Dokania
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 |
|---|---|---|---|
| fit post-hoc calibrator on in-distribution val set instead of domain-shifted Objects365 (ablation) | Isotonic Regression on D-DETR · COCO minitest | D-ECE: 14.2 -> 1.3 | Improved |
| train D-DETR on Cityscapes with 4x longer schedule and larger images (ablation) | D-DETR vs Cal-DETR baselines · Cityscapes | D-ECE: 13.8 -> 2.9 | Improved |
| replace train-time calibration (Cal-DETR) with post-hoc Isotonic Regression | D-DETR · COCO minitest | LaECE0: 11.6 -> 7.7 | Improved |
| replace train-time calibration (Cal-DETR) with post-hoc Isotonic Regression | D-DETR · Cityscapes minitest | LaECE0: 21.3 -> 9.0 | Improved |
| post-hoc calibration under large natural domain shift | Isotonic Regression on D-DETR · Obj45K | LaECE0: 16.3 -> 17.2 | Hurt |
| use domain-shifted val set for TS thresholds and calibrator (ablation) | Temperature Scaling · COCO minitest | LRP: 57.3 -> 61.5 | Hurt |
| add detection thresholding and class-wise calibrators to Temperature Scaling (ablation) | Temperature Scaling · Cityscapes minitest | LaECE0: 20.9 -> 12.7 | Improved |
| add bias term (Platt Scaling instead of TS) (ablation) | Temperature Scaling · Cityscapes minitest | LaECE0: 12.7 -> 9.6 | Improved |
| optimize calibrator for the evaluation measure (D-ECE vs LaECE objective) (ablation) | Platt Scaling · COCO minitest | LaECE: PS for D-ECE: LaECE 16.3 vs PS for LaECE: 11.5 | Mixed |
| calibrate 14 diverse detectors with Isotonic Regression | various detectors · COCO minitest | LaECE0: 10.6-34.4 -> 6.7-8.9 | Improved |
| use Isotonic Regression instead of Platt Scaling (ablation) | post-hoc calibrators · COCO minitest | LaECE0: 9.6 -> 7.7 | Improved |
| apply Cal-DETR style train-time regularization (MbLS) for calibration | D-DETR · COCO minitest | D-ECE: 12.8 -> 15.6 | Hurt |
Closest papers by meaning
- Rethinking Detection Calibration: A Coordinate Perspective (ECCV 2026 · similarity 0.90)
- Multiclass Confidence and Localization Calibration for Object Detection (CVPR 2023 · similarity 0.89)
- Systematic Evaluation of Uncertainty Calibration in Pretrained Object Detectors (IJCV 2024 · similarity 0.89)
- Bridging Precision and Confidence: A Train-Time Loss for Calibrating Object Detection (CVPR 2023 · similarity 0.89)
- Uncertainty Quantification in Detection Transformers: Object-Level Calibration and Image-Level Reliability (TPAMI 2026 · similarity 0.89)
- YOLOv10: Real-Time End-to-End Object Detection (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.
