Dataset · vidore
DocVQA test subset for document retrieval
vidore/docvqa_test_subsampled
A subset of 500 English question-answer pairs for document scans to evaluate multimodal retrieval and VQA.
What's inside
The dataset contains 500 English question-answer pairs linked to document images from the UCSF Industry Documents Library. Questions and answers are manually labeled, and the set is a reduced version of the original DocVQA test split with standardized column names.
What you can do with it
- Evaluating the quality of multimodal RAG systems and visual document retrieval
- Testing VLM models on answer extraction tasks from scans without prior OCR
- Tasks
- Document question answering, Visual document retrieval
- Languages
- English
- Size
- up to 1K rows
- Downloads this month
- 1.6K
- Likes
- 7
How to load it
from datasets import load_dataset
ds = load_dataset("vidore/docvqa_test_subsampled")