Open Sora Dataset Project Video Dataset

Open-Sora-Plan is an open source project that aims to reproduce OpenAI's Sora (T2V model) and build knowledge about Video-VQVAE (VideoGPT) + DiT. The project was jointly initiated by Peking University and Tuzhan Intelligence Company, and the research significantly enhanced the quality of video generation and the ability to control text. The model is able to generate 10 seconds, 24FPS 1024×1024 HD video, and also supports the generation of high-resolution images, providing users with a richer and more sophisticated visual experience.

This dataset is a video dataset for their project. The research team grabbed 40,258 videos from open source websites under the CC0 license. All videos are high-quality and watermark-free, of which about 60% are landscape data. The total length is about 274h 05m 13s .

The main sources of data are divided into three parts:

1. mixkit：The total number of videos collected by the research team is 1,234, the total duration is about 6h 19m 32s, the total number of frames is 570,815 .
2. pexels: The total number of videos collected by the research team is 7,408,Total duration is approximately 48h 49m 24s, the total number of frames is 5,038,641 .
3. pixabay: The total number of videos collected by the research team is 31,616,Total duration is approximately 218h 56m 17s, the total number of frames is 23,508,970 .