Research map / 3D vision / Human pose estimation
Learning to Reconstruct 3D Human Pose and Shape via Model-Fitting in the Loop
Nikos Kolotouros, Georgios Pavlakos, Michael J. Black, Kostas Daniilidis
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 fitting in the training loop instead of static SMPLify fits (ablation) | SPIN · 3DPW | reconstruction error (mm): 66.3 -> 59.2 | Improved |
| add model-based supervision from static fits versus prior HMR-style 2D losses | SPIN static fits · 3DPW | reconstruction error (mm): 70.2 -> 66.3 | Improved |
| add fitting in the training loop (ablation) | SPIN · LSP foreground-background | accuracy: 91.07 -> 91.83 | Improved |
| add fitting in the training loop (ablation) | SPIN · LSP part segmentation | accuracy: 88.48 -> 89.41 | Improved |
| add fitting in the training loop (ablation) | SPIN · LSP part segmentation | f1: 0.65 -> 0.68 | Improved |
| add fitting in the training loop (ablation) | SPIN · LSP foreground-background | f1: 0.86 -> 0.87 | Improved |
| use SPIN instead of HMR with 2D reprojection loss, with 3D ground truth | SPIN · Human3.6M | reconstruction error (mm): 56.8 -> 41.1 | Improved |
| use SPIN instead of HMR, unpaired (no 3D ground truth) | SPIN · Human3.6M | reconstruction error (mm): 66.5 -> 62.0 | Improved |
| use SPIN instead of HMR, unpaired | SPIN · MPI-INF-3DHP absolute | MPJPE: 169.5 -> 124.8 | Improved |
| use SPIN instead of HMR, with 3D ground truth | SPIN · MPI-INF-3DHP rigid alignment | MPJPE: 89.8 -> 67.5 | Improved |
| use SPIN instead of VNect, with 3D ground truth | SPIN · MPI-INF-3DHP absolute | PCK: 76.6 -> 76.4 | Hurt |
| use SPIN instead of SMPLify oracle at test time | SPIN · LSP foreground-background | accuracy: 92.17 -> 91.83 | Hurt |
Closest papers by meaning
- Human Body Model Fitting by Learned Gradient Descent (ECCV 2020 · similarity 0.92)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · similarity 0.92)
- Keep It SMPL: Automatic Estimation of 3D Human Pose and Shape from a Single Image (ECCV 2016 · similarity 0.90)
- 3D Human Pose Estimation From a Single Image via Distance Matrix Regression (CVPR 2017 · similarity 0.90)
- STAR: Sparse Trained Articulated Human Body Regressor (ECCV 2020 · similarity 0.90)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.89)
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.
