DocVQA

Fits agents that read scanned or photographed documents (forms, letters, reports) and pull out a short answer. Borrow its question-over-image case format and its ANLS scoring, which tolerates small OCR spelling errors instead of demanding an exact string.

What it asks the agent to do

• answer a question about a scanned document page from its image
• read a value out of a form, table or letter
• use the page layout to find where an answer sits

What to borrow from it

baselinecase template

Caveats

From 2020 and older than most of this index, though still widely reported; human accuracy is 94% and current vision models are close to it, test-split answers are held back, and the dataset license is not stated.

Details

Links

• Paper
• Leaderboard