Table 12 Comparison of 8-head and 12-head attention mechanism
From: BPI-MVQA: a bi-branch model for medical visual question answering
model | VQA-Med2018 | VQA-Med2019 | VQA-RAD | |||
|---|---|---|---|---|---|---|
W. | B. | A. | B. | A. | W. | |
BPI-MVQA with 8-head self-attention | 0.184 | 0.160 | 0.640 | 0.644 | 0.692 | 0.753 |
BPI-MVQA with 12-head self-attention | 0.188 | 0.162 | 0.654 | 0.687 | 0.636 | 0.737 |