Research map / Learning theory and algorithms / Prediction and decision losses
Adaptive Bounding Box Uncertainties via Two-Step Conformal Prediction
Alexander Timans, Christoph-Nikolas Straehle, Kaspar Sakmann, Eric Nalisnick
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 |
|---|---|---|---|
| use two-step conformal prediction (class-label ClassThr then box intervals) with alpha_L=0.01, alpha_B=0.1 | Faster R-CNN on COCO, Cityscapes, BDD100k · COCO, Cityscapes, BDD100k | label and box coverage | Improved |
| use conformal Box-Std intervals instead of deep ensembles or GaussianYOLO uncertainties | object detectors · COCO two-sided intervals, target 0.9 | coverage: 0.21 -> 0.88 (vs DeepEns); 0.08 -> 0.88 (vs GaussianYOLO) | Improved |
| use Box-Std one-sided intervals instead of Andeol et al. best | Faster R-CNN · COCO one-sided intervals | MPIW (pixels): 87.62 -> 85.42 | Improved |
| use adaptive Box-Ens or Box-CQR instead of fixed-width Box-Std (ablation) | bounding box intervals · BDD100k, coverage stratified by object size | coverage balance across sizes | Improved |
| use max-rank multiple testing correction instead of Bonferroni (ablation) | box coordinate intervals · object detection prediction intervals | interval width | Improved |
| use ClassThr label sets instead of the Full label set (ablation) | two-step pipeline · COCO, Cityscapes, BDD100k | box interval width | Improved |
| use Top singleton label set instead of ClassThr (ablation) | two-step pipeline · COCO, Cityscapes, BDD100k | label and box coverage | Hurt |
| use Naive density-level-set label sets instead of ClassThr (ablation) | two-step pipeline · COCO, Cityscapes, BDD100k | box coverage | Mixed |
Closest papers by meaning
- Bounding Box Regression With Uncertainty for Accurate Object Detection (CVPR 2019 · similarity 0.88)
- Bounding Box Stability against Feature Dropout Reflects Detector Generalization across Environments (ICLR 2024 · similarity 0.85)
- Optimal transport-based conformal prediction (ICML 2025 · similarity 0.85)
- OpenBox: Annotate Any Bounding Boxes in 3D (NeurIPS 2025 · similarity 0.84)
- Learning Polyhedral Conformal Sets for Robust Optimization (NeurIPS 2026 · similarity 0.84)
- Conformal Prediction and MLLM aided Uncertainty Quantification in Scene Graph Generation (CVPR 2025 · similarity 0.83)
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.
