Research map / 3D vision / Human pose estimation
Convolutional Mesh Regression for Single-Image Human Shape Reconstruction
Check your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Regress mesh vertices with a Graph-CNN rather than regress SMPL parameters directly | Human shape reconstruction · Human3.6M Protocol 2 | Reconstruction error (mm): 77.6 -> 61.3 | Helped |
| Use graph convolutions instead of fully connected layers for vertex regression (ablation) | Mesh regression · Human3.6M Protocol 2 | MPJPE (mm): 200.8 -> 102.1 | Helped |
| Use graph convolutions instead of fully connected layers for vertex regression (ablation) | Mesh regression · Human3.6M Protocol 2 | Reconstruction error (mm): 105.8 -> 69.0 | Helped |
| Regress SMPL parameters from the predicted shape (ablation) | RGB-based graph mesh regression · Human3.6M Protocol 2 | MPJPE (mm): 72.7 -> 71.9 | Helped |
| Regress SMPL parameters from the predicted shape (ablation) | Part-segmentation graph mesh regression · Human3.6M Protocol 2 | MPJPE (mm): 73.4 -> 77.4 | Hurt |
| Regress SMPL parameters from the predicted shape (ablation) | DensePose graph mesh regression · Human3.6M Protocol 2 | MPJPE (mm): 72.3 -> 74.2 | Hurt |
| Use DensePose features instead of RGB features (ablation) | Graph + SMPL mesh regression · Human3.6M Protocol 2 | Reconstruction error (mm): 50.1 vs 51.0 | Mixed |
| Use part segmentation instead of RGB features (ablation) | Graph + SMPL mesh regression · Human3.6M Protocol 2 | Reconstruction error (mm): 50.1 -> 53.3 | Hurt |
| Regress rotation matrices and project them to valid rotations instead of predicting axis-angle pose parameters (ablation) | SMPL parameter prediction from regressed shape · Human3.6M | Quantitative results | No clear effect |
| Train the shape network and SMPL-parameter MLP end-to-end instead of in two stages (ablation) | Graph mesh regression and SMPL parameter prediction · Human3.6M and UP-3D | Performance | Hurt |
| Use more powerful graph convolutions instead of the simpler adjacency-averaging convolution (ablation) | Graph mesh regression · Human3.6M | Quantitative results | No clear effect |
Closest papers by meaning
- Learning to Estimate 3D Human Pose and Shape From a Single Color Image (CVPR 2018 · similarity 0.93)
- End-to-End Recovery of Human Shape and Pose (CVPR 2018 · similarity 0.91)
- Coarse-To-Fine Volumetric Prediction for Single-Image 3D Human Pose (CVPR 2017 · similarity 0.90)
- Neural Localizer Fields for Continuous 3D Human Pose and Shape Estimation (NeurIPS 2024 · similarity 0.90)
- TokenHMR: Advancing Human Mesh Recovery with a Tokenized Pose Representation (CVPR 2024 · similarity 0.89)
- BLADE: Single-view Body Mesh Estimation through Accurate Depth Estimation (CVPR 2025 · 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.