MobileNetV2 has introduced significant changes in the architecture of MobileNet. Were the changes worth making? How much better is MobileNetV2 than MobileNet in regards to performance? We can compare the models in terms of the number of multiplication operations required for one inference, which is commonly known as MACs (number of multiply-accumulates). The higher the MAC value, the heavier the network is. We can also compare the models in terms of the number of parameters in the model. The following table shows the MACs and the number of parameters for both MobileNet and MobileNetV2:
| Network | Number of Parameters | MACs/ MAdds |
| MobileNet V1 | 4.2M | 575M |
| MobileNet V2 | 3.4M | 300M |
We can also compare the models in terms of memory that's required for the different number of channels and resolution. The following table provides...
