Research map / 3D vision / Human pose estimation
Weakly Supervised 3D Human Pose and Shape Reconstruction with Normalizing Flows
Andrei Zanfir, Eduard Gabriel Bazavan, Hongyi Xu, William T. Freeman, Rahul Sukthankar, Cristian Sminchisescu
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 body part alignment (BA) loss to keypoint alignment in optimization (ablation) | FS+OPT · 3DPW test (static) | MPJPE-PA (mm): 60.3 -> 58.87 | Improved |
| add body part alignment (BA) loss to keypoint alignment in optimization (ablation) | FS+OPT · 3DPW test (static) | MPJPE (mm): 95 -> 91.4 | Improved |
| add weakly supervised refinement on COCO after full supervision | FS+OPT (KA+BA) · 3DPW test (static) | MPJPE-PA (mm): 58.87 -> 57.1 | Improved |
| add keypoint-only optimization on top of the fully supervised model (ablation) | FS · 3DPW test (static) | MPJPE (mm): 95 -> 95 | No clear effect |
| add weakly supervised losses with only 20% of labeled H80K (ablation) | FS · H80K test | MPJPE (mm): 117 -> 97 | Improved |
| add weakly supervised losses starting from random initialization (0% supervision) (ablation) | FS · H80K test | MPJPE (mm): 649 -> 123 | Improved |
| add weakly supervised losses with 100% of labeled H80K (ablation) | FS · H80K test | MPJPE (mm): 83 -> 84 | No clear effect |
| increase temporal window from 4 to 16 frames in video optimization (ablation) | FS+WS+OPT(KA+BA+S) · 3DPW test (video) | MPJPE (mm): 84.5 -> 80.2 | Improved |
| add weakly supervised refinement before 16-frame smoothed optimization (ablation) | FS+OPT(KA+BA+S, 16 frames) · 3DPW test (video) | MPJPE-PA (mm): 52.2 -> 49.8 | Improved |
| add KA+BA instead of KA only during self-supervised refinement with 100% data (ablation) | WS · 3DPW test (COCO+OpenImages) | mIOU: 44.0 -> 50.0 | Improved |
| increase self-supervised data from 10% to 100% with KA+BA (ablation) | WS+KA+BA · 3DPW test (COCO+OpenImages) | 2D joint error (pixels): 7.2 -> 5.8 | Improved |
| add weakly supervised refinement to the fully supervised network (ablation) | FS · 3DPW test (COCO+OpenImages) | 2D joint error (pixels): 9.13 -> 5.8 | Improved |
Closest papers by meaning
- Self-Supervised 3D Human Pose Estimation via Part Guided Novel Image Synthesis (CVPR 2020 · similarity 0.91)
- Probabilistic Monocular 3D Human Pose Estimation With Normalizing Flows (ICCV 2021 · similarity 0.90)
- Learning 3D Human Pose from Structure and Motion (ECCV 2018 · similarity 0.90)
- HuManiFlow: Ancestor-Conditioned Normalising Flows on SO(3) Manifolds for Human Pose and Shape Distribution Estimation (CVPR 2023 · similarity 0.90)
- Self-Supervised 3D Human Mesh Recovery from a Single Image with Uncertainty-Aware Learning (AAAI 2024 · similarity 0.89)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.88)
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.
