OpenVid-1M
High-quality, large-scale text-to-video video dataset designed for researching and training generative video models. Openvid-1M offers videos of at least 512×512 pixels, with an OpenVidHD subset offering 1080p videos. The dataset includes carefully annotated text-video pairs, making it easy to directly train or fine-tune video quality. Uploading requires significant space and the use of scripts to manage large archives.
Over 1 million text-to-video videos in HD video format (≥512×512), including an OpenVidHD subset of 433K 1080p videos. Large ZIP files to decompress, CSV files for text-video pairs. Total size approximately several terabytes (OpenVidHD ~4.5 TB).
CC-BY-4.0 (videos come from public sources with their respective licenses, to be respected when using)
Description
OpenVid-1M is a high-quality text-to-video dataset containing over 1 million videos with a minimum resolution of 512×512 pixels. A subset called OpenVidHD brings together 433,000 videos in 1080p resolution for tasks that require better visual definition. The data is accompanied by CSV files that associate each video with its descriptive text, facilitating supervised learning and conditional video generation.
What is this dataset for?
- Train Conditioned Video Generation Models on Text
- Improve the quality and resolution of videos generated through fine-tuning
- Research on video architectures with realistic data and high resolution
Can it be enriched or improved?
Yes, the dataset can be supplemented by other video collections, or enhanced by fine annotation of metadata, descriptions, or contexts. Scripts are provided to facilitate the manipulation and recomposition of large files.
🔎 In summary
🧠 Recommended for
- Computer Vision Researchers
- Generative Video Model Developers
- Multimedia AI research laboratories
🔧 Compatible tools
- PyTorch
- TensorFlow
- FFmpeg
- Video annotation tools
💡 Tip
Use the provided scripts to effectively manage large files and optimize video preprocessing.
Frequently Asked Questions
What is the approximate size of the OpenVid-1M dataset?
The complete dataset is very large, numbering over 1 million videos, with the OpenVidHD subset alone taking up about 4.5TB of disk space.
What are the minimum resolutions of the videos in OpenVid-1M?
All videos have a minimum resolution of 512×512 pixels, while the OpenVidHD subset offers 1080p high definition videos.
Can I use this dataset for commercial purposes?
Yes, the CC-BY-4.0 license allows commercial use provided that the license and the rights of the original video sources included in the dataset are respected.
