Research map / 3D vision / Human pose estimation
UPose3D: Uncertainty-Aware 3D Human Pose Estimation with Cross-View and Temporal Cues
Vandad Davoodnia, Saeed Ghorbani, Marc-André Carbonneau, Alexandre Messier, Ali Etemad
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 pose compiler (cross-view and temporal refinement) (ablation) | UPose3D · Human3.6m | MPJPE (mm): 37.14 -> 26.42 | Improved |
| keep the original 2D image-branch predictions in the final estimate (ablation) | UPose3D · Human3.6m | MPJPE (mm): 69.90 -> 26.42 | Improved |
| add uncertainty modeling to both branches (normalizing flows) (ablation) | UPose3D · Human3.6m | MPJPE (mm): 48.11 -> 26.42 | Improved |
| add uncertainty to the pose compiler at inference only (ablation) | UPose3D · Human3.6m | MPJPE (mm): 26.42 -> 26.42 (PA-MPJPE 23.58 -> 23.42) | No clear effect |
| add image-branch uncertainty (ablation) | UPose3D · Human3.6m | MPJPE (mm): 27.61 -> 26.42 | Improved |
| extend temporal window from 1 to 27 frames (ablation) | UPose3D · Human3.6m | MPJPE (mm): 26.9 -> 26.4 | Improved |
| extend temporal window to 243 frames (ablation) | UPose3D · Human3.6m | MPJPE (mm): 26.42 -> 33.17 | Hurt |
| use criss-cross attention instead of full attention (ablation) | spatiotemporal encoder · Human3.6m (T=243) | MPJPE (mm): 34.97 -> 33.17 | Improved |
| use DLT initialization instead of zero initialization (ablation) | 3D keypoint MLE · Human3.6m | MPJPE (mm): 28.51 -> 26.42 | Improved |
| append epipolar line or relative camera position to point cloud inputs (ablation) | pose compiler · Human3.6m | MPJPE (mm): 26.42 -> 26.46 (epipolar line), 26.37 (camera pos.) | No clear effect |
| train on synthetic multi-view AMASS data and apply to unseen RICH environment | UPose3D · RICH (OoD) | MPJPE (mm): 66.0 -> 36.2 (vs HRNet-W48 + DLT) | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- AdaFuse — AdaFuse: Adaptive Multiview Fusion for Accurate Human Pose Estimation in the Wild (IJCV 2020) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- CoDePose: Multi-View 3D Human Pose Estimation via Coupled 2D-3D Denoising Diffusion (ECCV 2026 · similarity 0.90)
- AdaptPose: Cross-Dataset Adaptation for 3D Human Pose Estimation by Learnable Motion Generation (CVPR 2022 · similarity 0.89)
- Sparse-view Pose Estimation and Reconstruction via Analysis by Generative Synthesis (NeurIPS 2024 · similarity 0.88)
- SVTformer: Spatial-View-Temporal Transformer for Multi-View 3D Human Pose Estimation (AAAI 2025 · similarity 0.87)
- ExtPose: Robust and Coherent Pose Estimation by Extending ViTs (ICML 2025 · similarity 0.87)
- FreeMan: Towards Benchmarking 3D Human Pose Estimation under Real-World Conditions (CVPR 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.
