I’ve been working on a neural network that is able to identify ball python morphs from pictures. The model was trained on ~35k images of the top 20 most common visual morphs (and their combinations) in the market. The result is decent but the model struggles to identify subtle morphs.
Looking for ideas to train the model in more specific ways such as to:
- Determine if morph X exists
- Distinguish morph X from morph Y
- Anything related would be nice
On the prediction accuracy:
50% of the time there is no mistake;
28% of the time there is 1 mistake;
14% of the time there are 2 mistakes;
5% of the time there are 3 mistakes;
Here are some predictions (higher value means higher confidence):
