Research map / 3D vision / Human pose estimation
Learned Vertex Descent: A New Direction for 3D Human Model Fitting
Enric Corona, Gerard Pons-Moll, Guillem Alenyà, Francesc Moreno-Noguer
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 |
|---|---|---|---|
| replace IP-Net with LVD for SMPL registration | SMPL registration · Body registration (SMPL) | Joint error [cm]: 3.60 -> 2.53 | Improved |
| replace IP-Net with LVD for SMPL registration | SMPL registration · Body registration (SMPL) | Vertex error [cm]: 5.03 -> 3.00 | Improved |
| replace no-correspondence fitting with LVD initialization | SMPL registration · Body registration (SMPL) | Joint error [cm]: 16.6 -> 2.53 | Improved |
| add iterative LVD optimization on top of the forward pass (ablation) | LVD · Body registration (SMPL) | Joint error [cm]: 5.89 -> 2.53 | Improved |
| add iterative LVD optimization on top of the forward pass (ablation) | LVD · Body registration (SMPL) | Vertex error [cm]: 6.27 -> 3.00 | Improved |
| use LVD instead of Sengupta on BUFF | single-view SMPL · BUFF single-view SMPL | V2V avg [mm]: 43.05 -> 39.08 | Improved |
| use LVD instead of FrankMocap on BUFF | single-view SMPL · BUFF single-view SMPL | V2S avg [mm]: 38.66 -> 37.87 | Improved |
| use LVD instead of Sengupta for back view (180 deg) | single-view SMPL · BUFF single-view SMPL | V2V at 180 deg [mm]: 40.11 -> 54.55 | Hurt |
| use LVD instead of IP-Net for MANO registration | hand registration · Hand registration (MANO) | Joint error [cm]: 1.44 -> .76 | Improved |
| use LVD instead of IP-Net for MANO registration | hand registration · Hand registration (MANO) | Scan-to-recons V2S [mm]: 4.08 -> 3.33 | Improved |
| use deeper MLP or positional encoding (ablation) | LVD · LVD architecture | performance | No clear effect |
| refine SMPL registration to SMPL+D | LVD · Body registration | Recons-to-scan V2V [mm]: 8.16 -> 1.14 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- LoopReg — LoopReg: Self-supervised Learning of Implicit Surface Correspondences, Pose and Shape for 3D Human Mesh Registration (NeurIPS 2020) · better on 3 of 3 shared results
- SPIN — SPIN: Structure-Preserving Inner Offset Network for Scene Text Recognition (AAAI 2021) · better on 2 of 2 shared results
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
- Human Body Model Fitting by Learned Gradient Descent (ECCV 2020 · similarity 0.93)
- Learning to Reconstruct 3D Human Pose and Shape via Model-Fitting in the Loop (ICCV 2019 · similarity 0.90)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · similarity 0.89)
- Learning To Regress Bodies From Images Using Differentiable Semantic Rendering (ICCV 2021 · similarity 0.89)
- Neural Descent for Visual 3D Human Pose and Shape (CVPR 2021 · similarity 0.89)
- Recovering 3D Human Mesh From Monocular Images: A Survey (TPAMI 2023 · 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.
