Research map / 3D vision / Human pose estimation
Plausible Uncertainties for Human Pose Regression
Lennart Bramlage, Michelle Karg, Cristóbal Curio
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 |
|---|---|---|---|
| add isotonic-regression recalibration to predicted uncertainties (ablation) | DER and MAP · MS-COCO aleatoric ECE (DER) | ECE: .204 -> .025 | Improved |
| use multivariate instead of univariate DER (ablation) | DER · MS-COCO aleatoric ECE (before recalibration) | ECE: .204 -> .065 | Improved |
| use multivariate instead of univariate DER for pose accuracy (ablation) | DER · H36M 3d pose | mpjpe: 73.7 -> 68.9 on H36M; 14.5 -> 28.2 on SIM | Mixed |
| replace DER with MAP+MC dropout for efficiency (ablation) | uncertainty estimation · training and inference speed | speed: DER up to three times faster | Hurt |
| use DER instead of MAP for pose accuracy | ResNet-50 pose regression · MS-COCO pck02 | pck02: 88.9 -> 89.2 | No clear effect |
| use DER with ResNet-50 compared with RLE | MS-COCO val · MS-COCO val | AP: 70.5 -> 67.8 | Hurt |
| inject Gaussian label noise into SIM training labels (ablation) | DER · SIM label noise | aleatoric uncertainty | Improved |
| increase wrist occluder size (ablation) | DER · Occlusion trials | epistemic uncertainty | Mixed |
| use weight decay 1e-4 for DER (ablation) | DER · calibration | calibration | Improved |
| couple scalar evidence parameters linearly in multivariate DER (ablation) | multivariate DER · calibration | calibration | Hurt |
| use heatmap-based IHPR instead of probabilistic regression head | pose estimation · H36M mpjpe | mpjpe: 73.7 -> 109.8 | Hurt |
Closest papers by meaning
- Learning in an Uncertain World: Representing Ambiguity Through Multiple Hypotheses (ICCV 2017 · similarity 0.89)
- On the Calibration of Human Pose Estimation (ICML 2024 · similarity 0.86)
- An Assessment of Human vs. Model Uncertainty in Soft-Label Learning and Calibration (NeurIPS 2026 · similarity 0.86)
- Benchmarking Uncertainty Disentanglement: Specialized Uncertainties for Specialized Tasks (NeurIPS 2024 · similarity 0.86)
- CUPS: Improving Human Pose-Shape Estimators with Conformalized Deep Uncertainty (ICML 2025 · similarity 0.86)
- MotionMap: Representing Multimodality in Human Pose Forecasting (CVPR 2025 · 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.
