Benchmark
DocVQA
Question answering over document images, combining reading with document-layout understanding.
Explore the original evidence
Explore the original evaluation, its methods, and reported findings.
Open evalCoverage
The original paper describes 50,000 questions over more than 12,000 document images.
Methodology
Evaluate answers to questions grounded in document images. The project now hosts multiple challenge editions; select a specific edition before reproducing rankings.