Image Classification Problem (One Shot Learning with Siamese Networks) in Python -- 2

I have a dataset consisting of 10,000 labeled images. Each image is it's own class. So 10,000 images / 10,000 classes. I will need an accurate model building, in Python (well commented and documented).
The model will use Siamese networks with Triplet loss. The final prediction will use Keypoint matching for final reranking (ORB).