Research map / 3D vision / Human pose estimation
GHUM & GHUML: Generative 3D Human Shape and Articulated Pose Models
Hongyi Xu, Eduard Gabriel Bazavan, Andrei Zanfir, William T. Freeman, Rahul Sukthankar, Cristian Sminchisescu
Check 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 |
|---|---|---|---|
| Use the low-resolution GHUML model instead of GHUM (ablation) | GHUM/GHUML · Caesar and GHS3D body, head, and hand | ICP error (mm): Caesar: 0.265 -> 0.465; body: 0.371 -> 0.725; head: 0.442 -> 0.519; hand: 0.164 -> 0.423 | Hurt |
| Use the low-resolution GHUML model instead of GHUM (ablation) | GHUM/GHUML · Caesar and GHS3D body, face, and hand | mean vertex-based Euclidean reconstruction error (mm): Caesar: 2.81 -> 3.27; body: 5.21 -> 6.32; face: 2.96 -> 3.28; hand: 2.22 -> 2.81 | Hurt |
| Use GHUML rather than GHUM (ablation) | GHUM/GHUML · GHS3D model evaluation | feed-forward evaluation speed: 2.0× faster | Improved |
| Use a 16-dim rather than a 64-dim body-shape latent representation (ablation) | GHUM body VAE · Body shape reconstruction | reconstruction error: 1.72× higher | Hurt |
| Use a 20-dimensional VAE instead of 96 linear PCA bases (ablation) | Facial expression model · Facial expression reconstruction | reconstruction error: similar | No clear effect |
| Use a 20-dimensional VAE instead of 96 linear PCA bases (ablation) | Facial expression model · Facial expression reconstruction | performance: 1.4× slower | Hurt |
| Use GHUM rather than SMPL skinning | GHUM and SMPL · GHS3D motion sequence, body excluding face and hands | mean reconstruction error (mm): 4.96 mm -> 4.23 mm; 0.76 mm lower error on average | Improved |
Closest papers by meaning
- imGHUM: Implicit Generative Models of 3D Human Shape and Articulated Pose (ICCV 2021 · similarity 0.89)
- Neural Descent for Visual 3D Human Pose and Shape (CVPR 2021 · similarity 0.88)
- gDNA: Towards Generative Detailed Neural Avatars (CVPR 2022 · similarity 0.87)
- ATLAS: Decoupling Skeletal and Shape Parameters for Expressive Parametric Human Modeling (ICCV 2025 · similarity 0.87)
- Expressive Body Capture: 3D Hands, Face, and Body From a Single Image (CVPR 2019 · similarity 0.87)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · 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.
