Skip to content

Dataset · FreedomIntelligence

Medical VQA based on PubMed articles

FreedomIntelligence/PubMedVision

Large set of English medical image-question-answer pairs from PubMed.

What's inside

PubMedVision is a large-scale medical VQA dataset. Image-text pairs are extracted from PubMed and reformatted using GPT-4V for improved quality. Original captions, a Chinese version, and body_part and modality annotations are added. The main language is English, with a metadata size of 1 to 10 million examples.

What you can do with it

  • Training medical multimodal VQA models
  • LoRA fine-tuning for medical image question answering
  • Evaluating models on medical images and texts
Tasks
Question answering, Text generation
Languages
English
Size
1M–10M rows
Downloads this month
2.4K
Likes
107

Models trained on it

How to load it

from datasets import load_dataset

ds = load_dataset("FreedomIntelligence/PubMedVision")