Research map / Detection and segmentation / CNN features
Systematic Evaluation of Uncertainty Calibration in Pretrained Object Detectors
Denis Huseljic, Marek Herde, Paul Hahn, Mehmet Müjde, Bernhard Sick
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 |
|---|---|---|---|
| avoid post-processing (NMS, thresholding) in Faster-RCNN | Faster-RCNN · COCO animals (in-distribution) | MCE: 0.140 -> 0.077 | Improved |
| avoid post-processing in Faster-RCNN | Faster-RCNN · COCO animals (in-distribution) | NLL: 0.549 -> 0.403 | Improved |
| avoid post-processing in Faster-RCNN | Faster-RCNN · COCO all (in-distribution) | MCE: 0.036 -> 0.011 | Improved |
| avoid post-processing in Faster-RCNN | Faster-RCNN · COCO all (in-distribution) | TCE: 0.087 -> 0.077 | Improved |
| avoid post-processing in Faster-RCNN | Faster-RCNN · COCO animals (in-distribution) | TCE: 0.061 -> 0.080 | Hurt |
| use set-based DETR instead of Faster-RCNN with post-processing | DETR vs Faster-RCNN · COCO animals (in-distribution) | NLL: 0.549 -> 0.120 | Improved |
| use DETR instead of Faster-RCNN with post-processing | DETR vs Faster-RCNN · COCO traffic (in-distribution) | MCE: 0.085 -> 0.059 | Improved |
| shift the sample distribution (COCO to Open Images) on the easier animals subset | DETR · COCO animals | mAP: 0.701 -> 0.718 | Improved |
| shift the sample distribution on the harder traffic subset | DETR · COCO traffic | MCE: 0.059 -> 0.099 | Hurt |
| shift the sample distribution on traffic | Faster-RCNN with post-processing · COCO traffic | NLL: 0.637 -> 0.878 | Hurt |
| use entropy histograms to detect distribution shift | all detectors · COCO traffic vs Open Images shifted | entropy histogram | No clear effect |
| use predicted probabilities (entropy) to detect OOD objects | DETR and Faster-RCNN · Open Images OOD | entropy histogram | Hurt |
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 0.92)
- Cal-DETR: Calibrated Detection Transformer (NeurIPS 2023 · similarity 0.91)
- Towards Improving Calibration in Object Detection Under Domain Shift (NeurIPS 2022 · similarity 0.90)
- Uncertainty Quantification in Detection Transformers: Object-Level Calibration and Image-Level Reliability (TPAMI 2026 · similarity 0.90)
- Unsupervised Object Detection with Theoretical Guarantees (NeurIPS 2024 · similarity 0.87)
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.
