Dataset · Salesforce
Large Visual Grounding Corpus by Salesforce
Salesforce/blip3-grounding-50m
10 to 100 million image-text pairs for precise word-to-object mapping in images.
What's inside
A huge multimodal English dataset from Salesforce containing 10 to 100 million image-text examples. It is designed to improve the ability of vision-language models (VLMs) to match semantic concepts and complex descriptions (e.g., "object to the left of the dog") with specific regions in an image.
What you can do with it
- Training VLMs for precise object retrieval via text descriptions (referring expressions)
- Improving object detection and semantic segmentation quality in multimodal systems
- Reducing visual hallucinations in models answering questions about images
- Languages
- English
- Size
- 10M–100M rows
- Downloads this month
- 827
- Likes
- 32
How to load it
from datasets import load_dataset
ds = load_dataset("Salesforce/blip3-grounding-50m")