Table 5 Comparison of Polysemantic Transformer (PoT) experiments at various stages with different k \(\times\) k grids.
From: End to end polysemantic cooperative mixed task trainer for UAV target detection
Polysemantic Transformer (PoT) k \(\times\) k grids | #Params | FLOPs | Acc. | |||
|---|---|---|---|---|---|---|
ResNet-2 | ResNet-3 | ResNet-4 | ResNet-5 | |||
3\(\times\)3 (*3) | 3\(\times\)3 (*4) | 3\(\times\)3 (*5) | 3\(\times\)3 (*3) | 21.3M | 3.5G | 33.63 |
3\(\times\)3 (*3) | 5\(\times\)5 (*4) | 3\(\times\)3 (*5) | 3\(\times\)3 (*3) | 22.1M | 3.5G | 33.15 |
3\(\times\)3 (*3) | 3\(\times\)3 (*4) | 5\(\times\)5 (*5) | 3\(\times\)3 (*3) | 23.2M | 3.8G | 33.01 |
3\(\times\)3 (*3) | 3\(\times\)3 (*4) | 5\(\times\)5 (*5) | 5\(\times\)5 (*3) | 24.1M | 3.9G | 32.18 |
5\(\times\)5 (*3) | 5\(\times\)5 (*4) | 5\(\times\)5 (*5) | 5\(\times\)5 (*3) | 25.4M | 4.1G | 31.79 |
