Skip to content

Dataset · Dev-Jahn

OpenVid-1M Video Dataset in WebDataset Format

Dev-Jahn/OpenVid-1M-wds

Repackaged OpenVid-1M dataset into .tar shards for fast streaming during video model training.

What's inside

A repackaged version of the original nkp37/OpenVid-1M corpus (ICLR 2025) containing between 1 and 10 million records with English text prompts and videos. The video content itself remains unchanged, but files are reassembled into the WebDataset format (~2 GB .tar shards) for efficient sequential reading via HTTP or NFS.

What you can do with it

  • Training text-to-video generation models
  • Training video classification and understanding models
  • Streaming video data for distributed training
Tasks
Text → video, Video classification
Languages
English
Size
1M–10M rows
Downloads this month
27.4K
Likes
2

How to load it

from datasets import load_dataset

ds = load_dataset("Dev-Jahn/OpenVid-1M-wds")