Research map / 3D vision / Human pose estimation
Deep Unsupervised 3D Human Body Reconstruction from a Sparse set of Landmarks
Meysam Madadi, Hugo Bertiche, Sérgio Escalera
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 attention model ATN with joint loss L_J (ablation) | DeepMurf · SURREAL val | surface error T_out (mm): 72.7 -> 64.6 | Improved |
| add attention model ATN (with L_J and L_unpose) (ablation) | Psi0 · SURREAL val | surface error T_out (mm): 92.9 -> 64.6 | Improved |
| add pose regularization L_phi (ablation) | DeepMurf · SURREAL val | surface error T_out (mm): 77.3 -> 64.6 | Improved |
| add shape regularization L_beta (ablation) | DeepMurf · SURREAL val | surface error T_out (mm): 80.4 -> 64.6 | Improved |
| use soft instead of hard landmark-to-surface assignment (ablation) | DeepMurf · SURREAL val | surface error T_out (mm): 71.1 -> 64.6 | Improved |
| add cascading block Psi1 (ablation) | DeepMurf · SURREAL val | surface error T_out (mm): 64.6 -> 47.2 | Improved |
| add inverse kinematic loss L_unpose on full data (ablation) | DeepMurf + Psi1 · SURREAL val | input joint error J_in (mm): 97.6 -> 34.7 | Improved |
| add inverse kinematic loss L_unpose on full data (ablation) | DeepMurf + Psi1 · SURREAL val | surface error T_out (mm): 47.6 -> 47.2 | No clear effect |
| add L_unpose when training on 256 samples (ablation) | DeepMurf · SURREAL val (256 training samples) | surface error T_out (mm): 136.5 -> 121.5 | Improved |
| add Procrustes rotation alignment preprocessing (ablation) | DeepMurf + Psi1 · SURREAL val | surface error T_out (mm): 47.2 -> 22.7 | Improved |
| add more cascade blocks beyond Psi1 (ablation) | DeepMurf + Psi1 · SURREAL val | surface error T_out (mm): improving the error around 1mm | No clear effect |
| raise missing landmark rate tau from 0.1 to 0.5 (ablation) | DAE + DeepMurf · SURREAL val | surface error T_out (mm): 68.3 -> 100.4 | Hurt |
Closest papers by meaning
- Deep Autoencoder for Combined Human Pose Estimation and Body Model Upscaling (ECCV 2018 · similarity 0.88)
- SHARP: Shape-Aware Reconstruction of People in Loose Clothing (IJCV 2022 · similarity 0.88)
- End-to-End Recovery of Human Shape and Pose (CVPR 2018 · similarity 0.86)
- Skeleton2Mesh: Kinematics Prior Injected Unsupervised Human Mesh Recovery (ICCV 2021 · similarity 0.86)
- MultiGO: Towards Multi-level Geometry Learning for Monocular 3D Textured Human Reconstruction (CVPR 2025 · similarity 0.86)
- Reconstructing Humans with a Biomechanically Accurate Skeleton (CVPR 2025 · similarity 0.85)
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.
