Research map / Video / Video generation
CogVideo: Large-scale Pretraining for Text-to-Video Generation via Transformers
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 hierarchical multi-frame-rate generation instead of 1-stage generation with N_overlap=1 (ablation) | CogVideo · Kinetics-600 | FVD: 137.13 -> 108.27 | Helped |
| Use hierarchical multi-frame-rate generation instead of 1-stage generation with N_overlap=2 (ablation) | CogVideo · Kinetics-600 | FVD: 120.82 -> 108.27 | Helped |
| Initialize from CogView2 rather than random initialization (ablation) | CogVideo · Kinetics-600 | FVD: 166.13 -> 108.27 | Helped |
| Add video pretraining after CogView2 initialization (ablation) | CogVideo · Kinetics-600 | FVD: 124.92 -> 108.27 | Helped |
| Use CogVideo rather than DIGAN | CogVideo · UCF-101 | FVD: 577 -> 626 | Hurt |
| Use CogVideo rather than TATS-base | CogVideo · UCF-101 | IS: 79.28 vs 50.46 | Hurt |
| Use CogVideo rather than TriVD-GAN-FP | CogVideo · Kinetics-600 | FVD: 25.74 vs 109.23 | Hurt |
| Use hierarchical generation rather than 1-stage generation (ablation) | CogVideo · UCF-101 human evaluation | semantic relevance | Helped |
| Use hierarchical generation rather than 1-stage generation (ablation) | CogVideo · UCF-101 human evaluation | motion realism | Helped |
| Use hierarchical generation rather than 1-stage generation (ablation) | CogVideo · UCF-101 human evaluation | texture quality | Helped |
Closest papers by meaning
- CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025 · similarity 0.93)
- Snap Video: Scaled Spatiotemporal Transformers for Text-to-Video Synthesis (CVPR 2024 · similarity 0.89)
- ControlVideo: Training-free Controllable Text-to-video Generation (ICLR 2024 · similarity 0.88)
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos (CVPR 2024 · similarity 0.88)
- Video-LaVIT: Unified Video-Language Pre-training with Decoupled Visual-Motional Tokenization (ICML 2024 · similarity 0.87)
- LinGen: Towards High-Resolution Minute-Length Text-to-Video Generation with Linear Computational Complexity (CVPR 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.