Design a CNN from scratch

Question

So currently I am working on a system the require a CNN But iam still new at this field, so I learned tflearn and I know how to use it, but now iam confused on how many layers of convolution layer do I need and what is the size of the filter of it I have 28 * 28 image set as input and I want to get 24 output classes so basically all I need is to know the convolution layer pool layer and full connected layer Any journal or formula could help

Answer 1

You don't have to invent something new in most cases, so I advice you to simply choose one of well-known CNN architectures . Just ajdust parameters and size of whole network for your particular case if it doesn't fit out of the box .