Research map / Detection and segmentation / CNN features
Estimating and Evaluating Regression Predictive Uncertainty in Deep Object Detectors
Ali Harakeh, Steven L. Waslander
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 |
|---|---|---|---|
| Train DETR bounding box distributions with the energy score instead of NLL (ablation) | DETR · COCO | mAP (%): 37.10 -> 39.96 | Improved |
| Train RetinaNet bounding box distributions with the energy score instead of NLL (ablation) | RetinaNet · COCO | mAP (%): 35.11 -> 36.02 | Improved |
| Train FasterRCNN bounding box distributions with the energy score instead of NLL (ablation) | FasterRCNN · COCO | mAP (%): 37.14 -> 37.32 | Improved |
| Train DETR bounding box distributions with the energy score instead of DMM (ablation) | DETR · COCO | mAP (%): 39.82 -> 39.96 | Improved |
| Train RetinaNet bounding box distributions with DMM instead of the energy score (ablation) | RetinaNet · COCO | mAP (%): 36.02 -> 36.28 | Improved |
| Train FasterRCNN bounding box distributions with the energy score instead of DMM (ablation) | FasterRCNN · COCO | mAP (%): 37.24 -> 37.32 | Improved |
| Train DETR bounding box distributions with the energy score instead of NLL (ablation) | DETR · COCO | CE (Reg): 0.0512 -> 0.0068 | Improved |
| Train RetinaNet bounding box distributions with the energy score instead of NLL (ablation) | RetinaNet · COCO | CE (Reg): 0.0242 -> 0.0059 | Improved |
| Train FasterRCNN bounding box distributions with the energy score instead of NLL (ablation) | FasterRCNN · COCO | CE (Reg): 0.0049 -> 0.0051 | Hurt |
| Train DETR or RetinaNet bounding box distributions with NLL instead of the energy score (ablation) | DETR and RetinaNet · COCO validation | Predictive entropy | Hurt |
| Use ad-hoc IOU-threshold regression target assignment rather than broad target support | RetinaNet and FasterRCNN variance networks · COCO validation | Uncertainty reliability across prediction errors | Hurt |
Closest papers by meaning
- Systematic Evaluation of Uncertainty Calibration in Pretrained Object Detectors (IJCV 2024 · similarity 0.89)
- Energy-Based Models for Deep Probabilistic Regression (ECCV 2020 · similarity 0.88)
- Simple and Scalable Predictive Uncertainty Estimation using Deep Ensembles (NeurIPS 2017 · similarity 0.88)
- Object Detection As Probabilistic Set Prediction (ECCV 2022 · similarity 0.87)
- SURE: SUrvey REcipes for building reliable and robust deep networks (CVPR 2024 · similarity 0.85)
- From Risk to Uncertainty: Generating Predictive Uncertainty Measures via Bayesian Estimation (ICLR 2025 · similarity 0.84)
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.
