Research map / Video / Video generation
MotionStreamer: Streaming Motion Generation via Diffusion-based Autoregressive Model in Causal Latent Space
Lixing Xiao, Shunlin Lu, Huaijin Pi, Ke Fan, Liang Pan, Yueer Zhou, Ziyong Feng, Xiaowei Zhou, Sida Peng, Jingbo Wang
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 continuous Causal TAE latents instead of VQ-VAE tokens (ablation) | MotionStreamer · HumanML3D test | FID: 13.226 -> 11.790 | Improved |
| use causal VAE (Causal TAE) instead of non-causal VAE (ablation) | MotionStreamer · HumanML3D test | FID: 19.902 -> 11.790 | Improved |
| use compressed KL-regularized latent instead of plain autoencoder (ablation) | MotionStreamer · HumanML3D test | FID: 43.828 -> 11.790 | Improved |
| add QK normalization in the AR transformer (ablation) | MotionStreamer · HumanML3D test | FID: 12.324 -> 11.790 | Improved |
| add Two-Forward training (ablation) | MotionStreamer · HumanML3D test | FID: 12.944 -> 11.790 | Improved |
| add diffusion head instead of MSE loss regression (ablation) | MotionStreamer · HumanML3D test | FID: 59.296 -> 11.790 | Improved |
| use T5-XXL text encoder instead of CLIP (ablation) | MotionStreamer · HumanML3D test | FID: 14.234 -> 11.790 | Improved |
| use root loss weight lambda 7.0 instead of 5.0 or 9.0 (ablation) | Causal TAE · HumanML3D test | reconstruction FID: 0.696 -> 0.661 | Improved |
| use Causal TAE with hidden size 1024 instead of 512 at latent dim 16 (ablation) | Causal TAE · HumanML3D test | generation FID: 14.096 -> 11.790 | Improved |
| increase latent dimension (larger dimension, less compression) (ablation) | Causal TAE · HumanML3D test | reconstruction vs generation | Mixed |
| use CFG scale 4.0 over no guidance (ablation) | MotionStreamer · HumanML3D test | generation quality | Improved |
| use continuous latent AR model instead of VQ-LLaMA discrete variant | MotionStreamer · BABEL subsequence | R@3: 0.383 -> 0.568 | Improved |
Closest papers by meaning
- Rethinking Diffusion for Text-Driven Human Motion Generation: Redundant Representations, Evaluation, and Masked Autoregression (CVPR 2025 · similarity 0.90)
- AMD: Autoregressive Motion Diffusion (AAAI 2024 · similarity 0.89)
- Generative Pre-trained Autoregressive Diffusion Transformer (NeurIPS 2025 · similarity 0.88)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.88)
- Human Motion Diffusion as a Generative Prior (ICLR 2024 · similarity 0.87)
- DartControl: A Diffusion-Based Autoregressive Motion Model for Real-Time Text-Driven Motion Control (ICLR 2025 · 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.
