Research map / Video / Video generation
DH-FaceVid-1K: A Large-Scale High-Quality Dataset for Face Video Generation
Donglin Di, He Feng, Wenzhang Sun, Yongjia Ma, Hao Li, Wei Chen, Lei Fan, Tonghua Su, Xun Yang
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 |
|---|---|---|---|
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | CogVideoX · Text-to-Video | FVD: 129.59 -> 98.01 | Improved |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | AnimateDiff · Text-to-Video | FVD: 174.59 -> 156.13 | Improved |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | CogVideoX · Image-to-Video | FVD: 123.25 -> 92.31 | Improved |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | SVD · Image-to-Video | FVD: 162.59 -> 132.92 | Improved |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | EasyAnimate · Text-to-Video | CLIP: 0.9274 -> 0.9240 | Hurt |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | I2VGen-XL · Image-to-Video | CLIP: 0.9067 -> 0.9013 | Hurt |
| train on DH-FaceVid-1K instead of CelebV-Text (ablation) | CogVideoX · Text-to-Video | CLIP: 0.9388 -> 0.9401 | Improved |
| increase training data from 100h to 1000h (ablation) | CogVideoX-5B · Text-to-Video | FVD: 237.52 -> 98.01 | Improved |
| increase training data from 600h to 1000h (ablation) | CogVideoX-2B · Text-to-Video | FVD: 145.14 -> 150.27 | Hurt |
| increase training data from 100h to 600h (ablation) | CogVideoX-2B · Text-to-Video | FVD: 215.06 -> 145.14 | Improved |
| increase training data from 600h to 1000h (ablation) | CogVideoX-2B · Image-to-Video | FVD: 135.14 -> 141.36 | Hurt |
| increase model size from 2B to 5B at 1000h (ablation) | CogVideoX · Image-to-Video | FVD: 141.36 -> 92.31 | Improved |
Closest papers by meaning
- OpenHumanVid: A Large-Scale High-Quality Dataset for Enhancing Human-Centric Video Generation (CVPR 2025 · similarity 0.89)
- MultiHuman-Testbench: Benchmarking Image Generation for Multiple Humans (NeurIPS 2025 · similarity 0.88)
- Limitations of Face Image Generation (AAAI 2024 · similarity 0.87)
- VLOGGER: Multimodal Diffusion for Embodied Avatar Synthesis (CVPR 2025 · similarity 0.87)
- Synergizing Motion and Appearance: Multi-Scale Compensatory Codebooks for Talking Head Video Generation (CVPR 2025 · similarity 0.87)
- AI-Face: A Million-Scale Demographically Annotated AI-Generated Face Dataset and Fairness Benchmark (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.
