ScreenAI: A Vision-Language Model for UI and Infographics Understanding

Benchmark Model Rank Results
visual-question-answering-vqa-on-infographicvqaScreenAI 5B (4.62 B params, w/ OCR)#13ANLS: 65.90
visual-question-answering-on-docvqa-testScreenAI 5B (4.62 B params, w/OCR)#46ANLS: 0.8988
chart-question-answering-on-chartqaScreenAI 5B (4.62 B params, w/ OCR)#791:1 Accuracy: 76.7