Dataset · Video-Reason
1 Million video clips for video reasoning training
Video-Reason/VBVR-Dataset
Large-scale training dataset of 1 million video clips with text prompts and frames for video reasoning tasks.
What's inside
A training split from the Very Big Video Reasoning (VBVR) package, including 1,000,000 English video clips created using 100 reasoning task generators (10,000 examples each). Each item contains a video clip, start and end frames, and a text prompt for logical inference.
What you can do with it
- Training vision-language models (VLMs) in logical analysis and video reasoning
- Training models to predict events between start and end frames
- Fine-tuning VideoQA systems on complex synthetic scenarios
- Tasks
- Video classification, Visual question answering, Video text → text
- Languages
- English
- Size
- 1M–10M rows
- Downloads this month
- 496
- Likes
- 58
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("Video-Reason/VBVR-Dataset")