Visual Question Answering on RETVQA 1.0 (test)
84.2AccuracyMI-BART
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| MI-BARTImage context source=Oracle Images2023.06 | 84.2 | 85.6 | 79.8 | |
| Image stitch MI-BARTImage context source=Oracle Images2023.06 | 78.2 | 74.7 | 70.7 | |
| MI-BARTImage context source=Retrieved Images2023.06 | 76.5 | 79.3 | 70.9 | |
| Image stitch MI-BARTImage context source=Retrieved Images2023.06 | 72.1 | 76.6 | 66.8 | |
| Aggregate VQAImage context source=Oracle Images2023.06 | 69.2 | 17.1 | 13 | |
| Aggregate VQAImage context source=Retrieved Images2023.06 | 66.6 | 16.2 | 11.9 | |
| VLPImage context source=Oracle Images2023.06 | 65.1 | 70.2 | 58.8 | |
| VLPImage context source=Retrieved Images2023.06 | 65.1 | 70.2 | 58.8 | |
| Question onlyImage context source=Oracle Images2023.06 | 62.4 | 15.3 | 10.4 | |
| Question onlyImage context source=Retrieved Images2023.06 | 62.4 | 15.3 | 10.4 | |
| Global popularityImage context source=Retrieved Images2023.06 | 36.2 | 16.7 | 9.8 | |
| Per-category popularityImage context source=Oracle Images2023.06 | 27.8 | 16 | 7.6 | |
| Per-category popularityImage context source=Retrieved Images2023.06 | 27.8 | 16 | 7.6 | |
| Global popularityImage context source=Oracle Images2023.06 | 27.4 | 14.5 | 7.9 |