Research map / Video / Video generation
MoGAN: Improving Motion Quality in Video Diffusion via Few-Step Motion Adversarial Post-Training
Haotian Xue, Qi Chen, Zhonghao Wang, Xun Huang, Eli Shechtman, Jinrong Xie, Yongxin Chen
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 |
|---|---|---|---|
| add DMD regularization loss (ablation) | MoGAN · Ablation (Table 2) | motion score: 0.674 -> 0.983 | Improved |
| add DMD regularization loss (ablation) | MoGAN · Ablation (Table 2) | dynamics degree: 0.35 -> 0.98 | Improved |
| add DMD regularization loss (ablation) | MoGAN · Ablation (Table 2) | aesthetic: 0.50 -> 0.57 | Improved |
| add R1 and R2 discriminator regularization (ablation) | MoGAN · Ablation (Table 2) | smoothness (%): 95.1 -> 98.5 | Improved |
| add R1 and R2 discriminator regularization (ablation) | MoGAN · Ablation (Table 2) | motion score: 0.905 -> 0.983 | Improved |
| compute GAN loss in optical-flow space instead of video space (ablation) | MoGAN · Ablation (Table 2) | dynamics degree: 0.85 -> 0.98 | Improved |
| compute GAN loss in optical-flow space instead of video space (ablation) | MoGAN · Ablation (Table 2) | image quality: 0.67 -> 0.66 | No clear effect |
| distill 50-step Wan into 3-step with DMD only | Wan2.1-T2V-1.3B · VBench | dynamics degree: 0.83 -> 0.73 | Hurt |
| add motion GAN post-training to DMD 3-step model | DMD-only · VBench | motion score: 0.859 -> 0.973 | Improved |
| add motion GAN post-training to DMD 3-step model | DMD-only · VideoJAM-Bench | motion score: 0.898 -> 0.983 | Improved |
| add motion GAN post-training to DMD 3-step model | DMD-only · VBench | dynamics degree: 0.73 -> 0.96 | Improved |
| add motion GAN post-training to DMD 3-step model | DMD-only · VBench | smoothness (%): 98.8 -> 98.6 | Mixed |
Closest papers by meaning
- Video Diffusion Models are Training-free Motion Interpreter and Controller (NeurIPS 2024 · similarity 0.89)
- OSV: One Step is Enough for High-Quality Image to Video Generation (CVPR 2025 · similarity 0.89)
- Motion Consistency Model: Accelerating Video Diffusion with Disentangled Motion-Appearance Distillation (NeurIPS 2024 · similarity 0.88)
- EDEN: Enhanced Diffusion for High-quality Large-motion Video Frame Interpolation (CVPR 2025 · similarity 0.88)
- FlowMo: Variance-Based Flow Guidance for Coherent Motion in Video Generation (NeurIPS 2025 · similarity 0.88)
- Identifying and Solving Conditional Image Leakage in Image-to-Video Diffusion Model (NeurIPS 2024 · 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.
