Dataset · TencentARC
Long videos with structured descriptions
TencentARC/MiraData
A large Tencent dataset with long video clips and detailed text descriptions for training video generation.
What's inside
It contains structured text descriptions and metadata for long video clips, designed for training Sora-level generative models. The volume ranges from 100,000 to 1 million entries, with English descriptions. The data is provided in tabular and text formats, collected by researchers from ARC Lab (Tencent PCG) and the Chinese University of Hong Kong.
What you can do with it
- Training and fine-tuning text-to-video and image-to-video models
- Training long video understanding and classification algorithms
- Building systems that generate detailed descriptions for video feeds
- Tasks
- Image → video, Text → image, Text → video, Video classification
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 1.1K
- Likes
- 43
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("TencentARC/MiraData")