Layout and Task Aware Instruction Prompt for Zero-shot Document Image Question Answering

Benchmark Model Rank Results
visual-question-answering-vqa-on-infographicvqaClaude + LATIN-Prompt#19ANLS: 54.51
visual-question-answering-vqa-on-infographicvqaGPT-3.5 + LATIN-Prompt#22ANLS: 48.98
visual-question-answering-on-docvqa-testGPT-4#50ANLS: 0.884
visual-question-answering-on-docvqa-testClaude + LATIN-Prompt#65ANLS: 0.8336
visual-question-answering-on-docvqa-testGPT-3.5 + LATIN-Prompt#66ANLS: 0.8255