Research map / 3D vision / Human pose estimation
imGHUM: Implicit Generative Models of 3D Human Shape and Articulated Pose
Thiemo Alldieck, Hongyi Xu, 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 Fourier input mapping (ablation) | single-part network · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.085 -> 0.052 | Improved |
| add Fourier input mapping (ablation) | imGHUM · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.095 -> 0.036 | Improved |
| use Swish activation instead of Softplus (ablation) | single-part network · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.070 -> 0.052 | Improved |
| use multi-part network (head, hands, body) instead of single-part (ablation) | imGHUM · GHUM samples test-set (1000 samples), hands | IoU: 0.818 -> 0.929 | Improved |
| use multi-part network instead of single-part (ablation) | imGHUM · GHUM samples test-set (1000 samples), head | IoU: 0.967 -> 0.976 | Improved |
| use multi-part network instead of deeper single-part (ablation) | imGHUM · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.058 -> 0.036 | Improved |
| add BCE inside/outside loss L_l (ablation) | imGHUM · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.051 -> 0.036 | Improved |
| add fine-tuning on scan registrations after GHUM-sample training (ablation) | imGHUM · GHUM registration test-set | Chamfer x10^-3: 0.071 -> 0.040 | Improved |
| add GHUM samples to scan-registration training (ablation) | imGHUM · GHUM samples test-set (1000 samples) | IoU: 0.834 -> 0.969 | Improved |
| use Euler angles instead of 6D rotation (ablation) | imGHUM · GHUM samples test-set (1000 samples) | Chamfer x10^-3: 0.044 -> 0.036 | No clear effect |
| use imGHUM instead of GHUM mesh | imGHUM vs GHUM · GHUM registration test-set | Chamfer x10^-3: 0.055 -> 0.040 | Improved |
| use imGHUM (smaller) instead of NASA | imGHUM · Single-subject multi-pose (NASA comparison) | IoU: 0.839 -> 0.962 | Improved |
Closest papers by meaning
- StructLDM: Structured Latent Diffusion for 3D Human Generation (ECCV 2024 · similarity 0.90)
- GHUM & GHUML: Generative 3D Human Shape and Articulated Pose Models (CVPR 2020 · similarity 0.89)
- AG3D: Learning to Generate 3D Avatars from 2D Image Collections (ICCV 2023 · similarity 0.88)
- Convolutional Mesh Regression for Single-Image Human Shape Reconstruction (CVPR 2019 · similarity 0.88)
- ChatPose: Chatting about 3D Human Pose (CVPR 2024 · similarity 0.87)
- SCULPT: Shape-Conditioned Unpaired Learning of Pose-dependent Clothed and Textured Human Meshes (CVPR 2024 · 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.
