So I’ve made my first fish classifier and its working ok with 44 types of fish. My question is about best practices for extending the number of labels that the model can recognize.
So, if I start off with Shark, Ray, Guppy, Damselfish what is the best way to continue with the model. What I’ve been doing upto now is to add Tiger Shark, Nurse Shark, Manta Ray, Eagle Ray, etc and then start all over from scratch. Do I have to do this ? or can I train one model and then add a new fish on top of the existing model using transfer learning.
If I look at all the types of fish, insect etc… its going to end up with 10,000 to maybe 100,000 or more labels to classify. Is this doable the way we’ve been taught so far ??
Should we break it into a separate models, ie. one just for sharks ??
Overall, what is the limit if any on the number of labels a neural net like dogs vs cats can handle.
Thanks,
Shaun