M3Grounder: Mask-Based Multi-Span and Multi-Granular Grounding for Document QA

Open paper

Benchmark	Model	Rank	Results
chart-question-answering-on-chartqa	M3Grounder-Q	–	1:1 Accuracy: 83.9
multimodal-table-fact-verification-on-tabfact-image-input	M3Grounder-Q	–	Accuracy: 60.4