1,
Is the middle depth-wide layer 3
318? The lecturer never writes it down, and I am now certain about this.
2, what is the down side of MobileNet? If none, then the algorithm should be adopted anywhere, not restricted to tight computational budgets.
Thanks in advance.
Hi,
I will try to give you my thoughts about your questions…
- At the slide it is giving a generic number n for the size, so n x n x 3, assuming three channels. Thus it is not trying to repeat the same example as for the previous example with V1. Maybe that is what confused you. To check all layers and sizes for different n values please go to the specific table at the paper.
- The main advantage of this architecture is that it is particularly suitable for mobile appli-
cations, but if you are not working with those platforms you could use any other architectures. The best is to compare results versus computational cost. Not always the one with the lowest computational cost is giving you the best results given your use-case…
Happy learning,
Rosa
