Research map / Video / Video generation
OpenVid++: A Large-scale High-quality Dataset for Text-to-video Generation
Kepan Nan, Tiehan Fan, Rui Xie, Penghao Zhou, Yilin Zhang, Xiang Li, Jian Yang, Zhenheng Yang, Ying Yu Tai
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 |
|---|---|---|---|
| replace STDiT with MVDiT (ablation) | T2V model, 256x256, OpenVid-0.4M · T2V benchmark | VQA_A: 11.11 -> 22.39 | Improved |
| increase resolution 256 to 512 for STDiT (ablation) | STDiT · T2V benchmark | VQA_A: 11.11 -> 65.15 | Improved |
| train MVDiT on OpenVid-1M instead of OpenVid-0.4M (ablation) | MVDiT-256 · T2V benchmark | VQA_A: 22.39 -> 24.87 | Improved |
| add multi-head cross-attention (MHCA) (ablation) | MVDiT-256 · T2V benchmark | VQA_A: 13.9 -> 22.39 | Improved |
| add scaling parameter alpha (ablation) | MVDiT-256 · T2V benchmark | VQA_A: 3.16 -> 22.39 | Improved |
| add clarity filtering to aesthetics, temporal and motion filters (ablation) | data pipeline · T2V benchmark (0.6M, 256x256) | VQA_A: 20.32 -> 30.26 | Improved |
| add clarity filtering to other three filters (ablation) | data pipeline · T2V benchmark (0.6M, 256x256) | Blip_bleu: 24.43 -> 23.43 | Hurt |
| train on OpenVid-1M instead of WebVid-10M | STDiT 256x256 · T2V benchmark | VQA_A: 13.40 -> 17.78 | Improved |
| train on OpenVid-1M instead of Panda-50M | STDiT 256x256 · T2V benchmark | VQA_T: 9.60 -> 12.98 | Improved |
| train on OpenVidHD-0.4M instead of Panda-50M-HD | STDiT 1024x1024 · T2V benchmark | VQA_A: 13.48 -> 73.46 | Improved |
| train on OpenVidHD-0.4M instead of WebVid-10M with super-resolution | STDiT 1024x1024 · T2V benchmark | Warping_error: 0.0137 -> 0.0052 | Improved |
| train restoration model on OpenVidHD-0.4M data instead of WebVid/YouTube data | I2VGen-XL video restoration · UDM10 | LPIPS: 0.2731 -> 0.2015 | Improved |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 2 on most shared results.
- CogVideoX-5B — CogVideoX: Text-to-Video Diffusion Models with An Expert Transformer (ICLR 2025) · better on 0 of 2 shared results
- Upscale-A-Video — Upscale-A-Video: Temporal-Consistent Diffusion Model for Real-World Video Super-Resolution (CVPR 2024) · 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
- OpenVid-1M: A Large-Scale High-Quality Dataset for Text-to-video Generation (ICLR 2025 · similarity 0.98)
- A Recipe for Scaling up Text-to-Video Generation with Text-free Videos (CVPR 2024 · similarity 0.92)
- LVD-2M: A Long-take Video Dataset with Temporally Dense Captions (NeurIPS 2024 · similarity 0.90)
- OpenS2V-Nexus: A Detailed Benchmark and Million-Scale Dataset for Subject-to-Video Generation (NeurIPS 2025 · similarity 0.89)
- Vivid-ZOO: Multi-View Video Generation with Diffusion Model (NeurIPS 2024 · similarity 0.89)
- VidProM: A Million-scale Real Prompt-Gallery Dataset for Text-to-Video Diffusion Models (NeurIPS 2024 · similarity 0.89)
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.
