Research map / Image generation / Fast diffusion sampling
EasyTune: Efficient Step-Aware Fine-Tuning for Diffusion-Based Motion Generation
Xiaofeng Tan, Wanjiang Weng, Haodong Lei, Hongsong 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 EasyTune step-wise reward optimization instead of trajectory-level DRaFT-50 | MLD fine-tuning · HumanML3D | FID: 0.197 -> 0.132 | Improved |
| use EasyTune instead of DRaFT-50 for memory | MLD fine-tuning · HumanML3D | peak memory (GB): 37.32 -> 22.10 | Improved |
| apply EasyTune to MLD | MLD · HumanML3D | FID: 0.473 -> 0.194 | Improved |
| apply EasyTune to MLD++ | MLD++ · HumanML3D | FID: 0.073 -> 0.073 | No clear effect |
| apply EasyTune to MotionDiffuse | MotionDiffuse · HumanML3D | R-Precision Top-1: 0.491 -> 0.488 | Mixed |
| apply EasyTune to MoDiffuse | MoDiffuse · KIT-ML | Diversity: 11.10 -> 10.63 | Hurt |
| add KL regularization to EasyTune (ablation) | EasyTune · HumanML3D | FID: 0.132 -> 0.172 | Hurt |
| add KL regularization to EasyTune (ablation) | EasyTune · HumanML3D | peak memory (GB): 22.10 -> 29.70 | Hurt |
| optimize all denoising steps instead of only final 20 steps (ablation) | EasyTune · HumanML3D | FID: 0.184 -> 0.132 | Improved |
| optimize all denoising steps instead of only initial 20 steps (ablation) | EasyTune · HumanML3D | FID: 0.158 -> 0.132 | Improved |
| use linearly decreasing step weights emphasizing early steps (ablation) | EasyTune · HumanML3D | R-Precision Top-1: 0.581 -> 0.584 | No clear effect |
| add SPL self-refining preference learning to ReAlign reward model (ablation) | ReAlign · HumanML3D text-motion retrieval | R@1: 67.59 -> 69.31 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- ReAlign — ReAlign: Generalizable Image Forgery Detection via Reasoning-Aligned Representation (CVPR 2026) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- AdaDiff: Adaptive Step Selection for Fast Diffusion Models (AAAI 2025 · similarity 0.89)
- Shortcut Diffusion Training with Cumulative Consistency Loss: An Optimal Control View (ICLR 2026 · similarity 0.89)
- MotionRFT: Unified Reinforcement Fine-Tuning for Text-to-Motion Generation (TPAMI 2026 · similarity 0.89)
- MotionMix: Weakly-Supervised Diffusion for Controllable Motion Generation (AAAI 2024 · similarity 0.88)
- Optimizing Diffusion Noise Can Serve As Universal Motion Priors (CVPR 2024 · similarity 0.88)
- SADA: Stability-guided Adaptive Diffusion Acceleration (ICML 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.
