Research map / 3D vision / Human pose estimation
ActFormer: A GAN-based Transformer towards General Action-Conditioned 3D Human Motion Generation
Liang Xu, Ziyang Song, Dongliang Wang, Jing Su, Zhicheng Fang, Chenjing Ding, Weihao Gan, Yichao Yan, Xin Jin, Xiaokang Yang, Wenjun Zeng, Wei Wu
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 |
|---|---|---|---|
| use Gaussian Process latent prior instead of Gaussian prior (ablation) | ActFormer · NTU-1P | FID_m: 17.25 -> 12.53 | Improved |
| use token-based Transformer generator instead of graph convolution with skeleton topology (ablation) | ActFormer · NTU-1P | FID_m: 18.68 -> 12.53 | Improved |
| use learnable instead of fixed positional encoding (ablation) | ActFormer · NTU-1P | FID_m: 18.38 -> 12.53 | Improved |
| use 2D temporal plus person-wise PE instead of fully independent PE (ablation) | ActFormer · NTU-2P | Acc.: 67.19 -> 69.65 | Improved |
| concatenation plus permutation augmentation in discriminator instead of pooling or self-attention (ablation) | ActFormer · NTU-2P | FID_m: 10.71 -> 3.77 | Improved |
| share one latent vector sequence across persons (ablation) | ActFormer · NTU-2P | FID_m: 14.79 -> 3.77 | Improved |
| add I-Former for explicit person interaction modeling instead of concatenation (ablation) | ActFormer · NTU-2P | FID_m: 7.63 -> 3.77 | Improved |
| add I-Former at 5 persons instead of concatenation (ablation) | ActFormer · GTA Combat 5 persons | FID_a: 7.62 -> 2.23 | Improved |
| independent PE versus full model on aggregated FID_a_m (ablation) | ActFormer · NTU-2P | FID_a_m: 7.00 -> 7.12 | Mixed |
Closest papers by meaning
- Action-Conditioned 3D Human Motion Synthesis With Transformer VAE (ICCV 2021 · similarity 0.89)
- Aggregated Multi-GANs for Controlled 3D Human Motion Prediction (AAAI 2021 · similarity 0.89)
- Scene-Aware Generative Network for Human Motion Synthesis (CVPR 2021 · similarity 0.89)
- MotionDreamer: One-to-Many Motion Synthesis with Localized Generative Masked Transformer (ICLR 2025 · similarity 0.86)
- FutureHuman3D: Forecasting Complex Long-Term 3D Human Behavior from Video Observations (CVPR 2024 · similarity 0.86)
- Motion Diversification Networks (CVPR 2024 · similarity 0.86)
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.
