UQ-DETR
Uncertainty Quantification in Detection Transformers: Object-Level Calibration and Image-Level Reliability
A lightweight Python toolkit for evaluating calibration and reliability of object detectors.
- Model-agnostic: Works with DETR, DINO, RT-DETR, Faster R-CNN, YOLO, or any custom detector.
- Minimal dependencies: Only
numpyandscipy. - Pip-installable:
pip install uq-detr.
What's in the package?
| Category | Functions |
|---|---|
| Calibration metrics | oce, dece, laece, lrp |
| Image-level reliability | contrastive_conf |
| Post-processing | select (threshold, top-k, NMS) |
| Matching | hungarian_match, compute_iou_matrix, compute_giou_matrix |
| Utilities | box_convert, Detections, GroundTruth |
Paper
Park, Sobolewski, and Azizan. "Uncertainty Quantification in Detection Transformers: Object-Level Calibration and Image-Level Reliability." IEEE Trans. Pattern Analysis and Machine Intelligence, 2025. arXiv:2412.01782