Research map / Video / Video generation
PoseGPT: Quantization-Based 3D Human Motion Generation and Forecasting
Thomas Lucas, Fabien Baradel, Philippe Weinzaepfel, Grégory Rogez
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 |
|---|---|---|---|
| PoseGPT instead of ACTOR | ACTOR · HumanAct12 | FID: 0.12 -> 0.08 | Improved |
| PoseGPT instead of ACTOR | ACTOR · BABEL | FID: 9.5 -> 2.7 | Improved |
| PoseGPT instead of ACTOR | ACTOR · GRAB | FID: 20.7 -> 5.1 | Improved |
| use 2 instead of 1 product-quantization codebooks (ablation) | PoseGPT · BABEL (T/Td=2, C=256) | FID: 3.5 -> 2.8 | Improved |
| use 8 instead of 4 codebooks (ablation) | PoseGPT · BABEL (T/Td=2, C=256) | FID: 2.7 -> 4.4 | Hurt |
| use 8 instead of 4 codebooks at temporal downsampling 4 (ablation) | PoseGPT · BABEL (T/Td=4, C=256) | FID: 3.1 -> 2.5 | Improved |
| embed the action at each timestep instead of as an extra input token (ablation) | PoseGPT · BABEL | sample quality | Improved |
| condition the GPT on sequence length (ablation) | PoseGPT · BABEL | sample quality | Improved |
| concatenate-and-project embeddings instead of summation (ablation) | PoseGPT · BABEL | sample quality | Mixed |
| use MLP head instead of single fully-connected layer (ablation) | PoseGPT · BABEL | sample quality | Improved |
| add auto-regressive prediction head over the K codebooks (ablation) | PoseGPT · BABEL | sample quality | Improved |
| add causal attention to the encoder (ablation) | PoseGPT · BABEL | sample quality | Hurt |
Closest papers by meaning
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.90)
- SimpliHuMoN: Simplifying Human Motion Prediction (NeurIPS 2026 · similarity 0.89)
- HumanDreamer: Generating Controllable Human-Motion Videos via Decoupled Generation (CVPR 2025 · similarity 0.89)
- Motion Diversification Networks (CVPR 2024 · similarity 0.89)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.88)
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.88)
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.
