ZFTurbo,

This is awesome. Great stuff?

Would you show us the code, or a resource online where we could learn to replicate this?

Much as you said you made it for fun, did it help you create a better model?

Thanks a lot

Visualisation was based on my old Kaggle code:

https://www.kaggle.com/code/zfturbo/visualisation-of-siamese-net

Difference is that here I used GlobalAveragePooling for NN and not GlobalMaxPooling. It's harder to visualize in this case. )

I made visualisation after contest. So it didn't help.

This is good interpretable ML. Seems the body markers are the main feature for recognition. Thank you for sharing this with the community Roman @ZFTurbo.