Why using same F3 filter for shortcut?

Course Q&A Deep Learning Specialization Convolutional Neural Networks
,
1

In the Ex2 of Residual_Networks, we use same F3 filter for shortcut path instead of using dedicated filter for shortcut path. Why is it OK?

4

Note that F3 just specifies the number of output channels at the given Conv2d layer, so it’s not actually an instantiated filter that’s being shared. The filters for each Conv2d layer are randomly initialized and will be trained individually by gradient descent. So the actual trained weight and bias values for the two filters with F3 output channels will likely be different.