Dataset · AudioVisual-Caption
Audio-visual video instructions
AudioVisual-Caption/ASID-1M
ASID-1M: a large English instruction dataset for video understanding with controlled descriptions.
What's inside
ASID-1M is a large English audio-visual instruction dataset for general video understanding. It provides fine-grained and controllable supervision: complex audio-visual content is described by more than a single general caption. The modality in metadata is text, the task is image-text-to-text, and the size is from 100K to 1M examples.
What you can do with it
- Training video-language models
- LoRA for controllable video descriptions
- Evaluating completeness of audio-visual captions
- Tasks
- Image text → text
- Languages
- English
- Size
- 100K–1M rows
- Downloads this month
- 1.1K
- Likes
- 85
How to load it
from datasets import load_dataset
ds = load_dataset("AudioVisual-Caption/ASID-1M")