Research map / 3D vision / Human pose estimation
Predicting 3D Human Dynamics From Video
Jason Y. Zhang, Panna Felsen, Angjoo Kanazawa, Jitendra Malik
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 |
|---|---|---|---|
| predict in the latent movie-strip space instead of 3D mesh space (ablation) | autoregressive predictor · Human3.6M, 30 frames (DTW) | reconstruction error (mm): 82.6 -> 65.1 | Improved |
| predict in the latent space instead of mesh space (ablation) | autoregressive predictor · Penn Action, 30 frames (DTW) | PCK: 56.5 -> 77.2 | Improved |
| predict in the latent space instead of mesh space (ablation) | autoregressive predictor · Human3.6M, 30 frames (no DTW) | reconstruction error (mm): 105.2 -> 81.1 | Improved |
| predict in the latent space instead of mesh space (ablation) | autoregressive predictor · Penn Action, 30 frames (no DTW) | PCK: 35.1 -> 60.1 | Improved |
| add the movie-strip distillation loss (ablation) | AR on latent Phi · Penn Action, 30 frames (DTW) | PCK: 75.6 -> 77.2 | Improved |
| add the movie-strip distillation loss (ablation) | AR on latent Phi · Human3.6M, 30 frames (DTW) | reconstruction error (mm): 65.3 -> 65.1 | No clear effect |
| autoregressive latent prediction instead of constant no-motion baseline | AR on latent Phi · Human3.6M, 30 frames (DTW) | reconstruction error (mm): 90.4 -> 65.1 | Improved |
| autoregressive latent prediction instead of nearest neighbor baseline | AR on latent Phi · Penn Action, 30 frames (DTW) | PCK: 57.8 -> 77.2 | Improved |
| use a causal temporal encoder without future frames instead of the bidirectional one | temporal encoder · Human3.6M | MPJPE (mm): 83.7 -> 83.9 | No clear effect |
| use the causal encoder | temporal encoder · Penn Action | PCK: 79.6 -> 80.6 | No clear effect |
| condition on a sequence of past frames in autoregressive prediction instead of a single image | future predictor · Penn Action, 5 frames ahead | PCK: 68.1 -> 77.2 | Improved |
Closest papers by meaning
- Learning 3D Human Dynamics From Video (CVPR 2019 · similarity 0.92)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.91)
- Forecasting Characteristic 3D Poses of Human Actions (CVPR 2022 · similarity 0.89)
- Existence Is Chaos: Enhancing 3D Human Motion Prediction with Uncertainty Consideration (AAAI 2024 · similarity 0.87)
- Harmonizing Stochasticity and Determinism: Scene-responsive Diverse Human Motion Prediction (NeurIPS 2024 · similarity 0.87)
- VideoMDM: Towards 3D Human Motion Generation From 2D Supervision (NeurIPS 2026 · 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.
