Research map / 3D vision / Human pose estimation
Weakly-supervised Learning of Human Dynamics
Petrissa Zell, Bodo Rosenhahn, Bastian Wandt
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 cyclic forward-dynamics training on the full motion-set to the supervised baseline (Dynamics Network) | Baseline net · Gait data set with 10% of ground-reaction and torque sets | joint torque RMSE and simulated-state RMSE | Improved |
| add the inverse layer and contact loss (cFI-net) to forward-only training (F-net) for domain transfer gait to run (ablation) | F-net · Domain transfer gait to run | GRF RMSE (N/kg): 3.942 -> 1.445 | Improved |
| add the inverse layer and contact loss (cFI-net) to forward-only training (F-net) for domain transfer run to gait (ablation) | F-net · Domain transfer run to gait | GRF RMSE (N/kg): 3.579 -> 0.685 | Improved |
| use cFI-net instead of F-net on the full gait set (ablation) | Dynamics Network · Gait data set | GRF RMSE (N/kg): 0.626 -> 0.733 | Hurt |
| use F-net instead of the supervised baseline net on the full gait set | Baseline net · Gait data set | joint torque RMSE (Nm/kg): 0.055 -> 0.053 | Improved |
| use F-net instead of the supervised baseline net on the full gait set | Baseline net · Gait data set | GRF RMSE (N/kg): 0.591 -> 0.626 | Hurt |
| increase angle noise from 0.3 to 2.3 deg (ablation) | F-net · Gait data set with angle noise | GRF RMSE (N/kg): 0.626 -> 1.096 | Hurt |
| increase angle noise from 0.3 to 2.3 deg (ablation) | cFI-net · Gait data set with angle noise | GRF RMSE (N/kg): 0.674 -> 1.143 | Hurt |
| train F-net additionally on CMU samples with the forward loss | F-net · CMU walking sequences (qualitative) | vertical GRF symmetry | Improved |
| add contact loss to inverse-loss training (ablation) | cFI-net · Domain transfer | GRF plausibility | Improved |
Closest papers by meaning
- Optimal-state Dynamics Estimation for Physics-based Human Motion Capture from Videos (NeurIPS 2024 · similarity 0.85)
- PhysPT: Physics-aware Pretrained Transformer for Estimating Human Dynamics from Monocular Videos (CVPR 2024 · similarity 0.85)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.84)
- ImDy: Human Inverse Dynamics from Imitated Observations (ICLR 2025 · similarity 0.84)
- Learning Physics From Video: Unsupervised Physical Parameter Estimation for Continuous Dynamical Systems (CVPR 2025 · similarity 0.84)
- H-MoRe: Learning Human-centric Motion Representation for Action Analysis (CVPR 2025 · similarity 0.83)
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.
