Research map / Video / Video generation
Learning to Generate Highly Dynamic Videos using Synthetic Motion Data
Wonjoon Jin, Jiyun Won, Janghyeok Han, Qi Dai, Chong Luo, Seung-Hwan Baek, Sunghyun Cho
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 synthetic motion data to motion generator training (ablation) | DynaVid · DynaVid-Human test | FVD: +527.04 FVD without synthetic data | Improved |
| add synthetic motion data to motion generator training (ablation) | DynaVid · Pexels | FVD: -49.85 FVD without synthetic data | No clear effect |
| mix real and synthetic flows in fine-tuning batches (ablation) | DynaVid motion generator · Pexels | FVD: +759.36 FVD without batch mixture | Improved |
| mix real and synthetic flows in fine-tuning batches (ablation) | DynaVid motion generator · DynaVid-Human test | FVD: -122.24 FVD without batch mixture | No clear effect |
| filter real motion-video pairs by flow cycle consistency (1.19 px threshold) (ablation) | DynaVid video generator · DynaVid-Human test | M-Err: 4.734 -> 4.287 | Improved |
| filter real motion-video pairs by flow cycle consistency (ablation) | DynaVid video generator · Pexels / DynaVid-Human | FVD, A-Qual: Pexels FVD -33.79 without filtering; DynaVid-Human +32.44 | No clear effect |
| train on rendered synthetic videos instead of synthetic motion (ablation) | DynaVid · Pexels | FVD: +104.43 FVD with rendered videos | Hurt |
| train on rendered synthetic videos instead of synthetic motion (ablation) | DynaVid · DynaVid-Human test | I-Qual: -0.0487 I-Qual; FVD -653.98 due to synthetic look | Hurt |
| DynaVid versus Wan2.2-5B on dynamic human motion | DynaVid · DynaVid-Human test | FVD: 1775.99 -> 1351.94 | Improved |
| DynaVid versus AC3D on rapid camera motion | DynaVid · DynaVid-Camera test | mRotErr: 1.1529 -> 0.9289 | Improved |
| DynaVid versus GEN3C on common camera motion | DynaVid · RE10K test | mRotErr: 0.0809 (GEN3C) vs 0.1136 | Hurt |
| add Gaussian noise to input flow (25 to 5 dB) (ablation) | motion-guided video generator · DynaVid-Human test | A-Qual: 0.5312 -> 0.4632 | Hurt |
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.
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 2 of 2 shared results
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
- DynaTokens: Teaching Dynamics to Camera-Controlled Video Models at Test Time (NeurIPS 2026 · similarity 0.89)
- Extrapolating and Decoupling Image-to-Video Generation Models: Motion Modeling is Easier Than You Think (CVPR 2025 · similarity 0.89)
- X-Dyna: Expressive Dynamic Human Image Animation (CVPR 2025 · similarity 0.89)
- Wan-Move: Motion-controllable Video Generation via Latent Trajectory Guidance (NeurIPS 2025 · similarity 0.89)
- FloVD: Optical Flow Meets Video Diffusion Model for Enhanced Camera-Controlled Video Synthesis (CVPR 2025 · similarity 0.89)
- MotionPro: A Precise Motion Controller for Image-to-Video Generation (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.
