For the above network, does the calculated RAM size include the input buffer size also? Its probably nit-picking analysis for the above case, but with larger networks, this inclusion/exclusion of input size could drastically change the numbers.
Good question, I believe it does, as if this was excluded it could lead to under estimation, that would cause memory allocation issues. It could vary for the more novel architectures, but I would need to check with the embedded team. Do you have a particular architecture or device in mind?
Do you want to have a call with one of our solutions / partner or other teams, or have you got a point of contact already? You can PM me some details, or I can loop in the relevant team here.