When I check out the MCUNetV1 code from github, I want to see how they delt with the Fully connected Layer (because imagenet has 1000 class and It has been written they have used at MCU of 256kb, but with 1000 class and connect those 1000 class with FCL(fully connected Layer), I don't think it would be over 256kb.)
So, is there anyone who can tell me how the architecture they have been using for Imagenet? (of course without actually running the code. I don't think my resource can afford to run whole MCUNet NAS Code.)
Thanks! I really appreciate with your answers and reply!