Table 6 Cost comparison between different variants.
From: Human motion recognition based on feature fusion and residual networks
Method | Parameters | FLOPs | Accuracy/% |
|---|---|---|---|
Textual model | 1.67 M | 1839.79 M | 99.00 |
No CBAM | 1.63 M | 1831.28 M | 98.33 |
Use a normal residual block | 11.47 M | 7303.13 M | 99.17 |
The fully connected layer has only one layer | 1.41 M | 1839.28 M | 97.67 |
