DePlot: One-shot visual language reasoning by plot-to-table translation

Open paper
Benchmark Model Rank Results
chart-question-answering-on-chartqaDePlot+FlanPaLM+Codex (PoT Self-Consistency)#21:1 Accuracy: 79.3
chart-question-answering-on-chartqaDePlot+Codex (PoT Self-Consistency)#31:1 Accuracy: 76.7
chart-question-answering-on-chartqaDePlot+FlanPaLM (Self-Consistency)#81:1 Accuracy: 70.5
chart-question-answering-on-chartqaDePlot+FlanPaLM (CoT)#111:1 Accuracy: 67.3
chart-question-answering-on-chartqaDePlot+GPT3 (Self-Consistency)#211:1 Accuracy: 42.3
chart-question-answering-on-chartqaDePlot+GPT3 (CoT)#221:1 Accuracy: 36.9
chart-question-answering-on-plotqaDePlot+FlanPaLM+Codex (PoT Self-Consistency)#21:1 Accuracy: 66.6
factual-inconsistency-detection-in-chart-1DePlot + GPT-4#2Kendall's Tau-c: 0.117
factual-inconsistency-detection-in-chart-2DePlot + GPT-4#3Kendall's Tau-c: 0.109
factual-inconsistency-detection-in-chart-3DePlot + GPT-4#2Kendall's Tau-c: 0.129

Rank counts only results with a code link.