Table 4 Chart QA experiment results.
From: Enterprise chart question and answer method based on multi modal cross fusion
Models | FigureQA | DVQA (OCR) | MECD | |||
|---|---|---|---|---|---|---|
Val1 | Val2 | Test-familiar | Test-novel | Val | test | |
IMG + QUES | 59.41% | 57.14% | 32.01% | 32.01% | 22.38% | 23.03% |
SANDY (Oracle) | 62.02% | 59.54% | 56.48% | 56.62% | 36.15% | 38.02% |
PReFIL | 94.84% | 93.26% | 80.88% | 80.04% | 49.43% | 51.46% |
VL-T5 | 83.59% | 82.38% | 85.78% | 84.47% | 54.89% | 56.31% |
Vision TaPas | 81.21% | 89.74% | 86.93% | 86.77% | 61.37% | 62.05% |
Pix2Struct | 89.83% | 88.62% | 89.78% | 90.13% | 77.26% | 76.95% |
OURS (Chart QA) | 91.37% | 91.45% | 94.43% | 94.37% | 80.56% | 82.24% |
