Dataset · pixparse
Single-page document questions and answers (DocVQA)
pixparse/docvqa-single-page-questions
A classic English benchmark of 50,000 questions on document images for evaluating Document AI.
What's inside
The popular DocVQA dataset (Mathew et al., 2021) contains 50,000 English questions compiled for over 12,000 images of various documents. The data is adapted for working with the Hugging Face datasets library.
What you can do with it
- Training and evaluating multimodal models for understanding scans and documents
- Benchmarking OCR + LLM pipelines for extracting facts from annotated text
- Fine-tuning models for document workflow automation
- Tasks
- Question answering
- Languages
- English
- Size
- 10K–100K rows
- Downloads this month
- 4.6K
- Likes
- 12
Models trained on it
How to load it
from datasets import load_dataset
ds = load_dataset("pixparse/docvqa-single-page-questions")