We have tried ORB a little, but we couldn't stop it from concentrating solely on the sand in the faces of the turtle or on the backgrounds.

I tried content-based algorithms like ORB/SIFT to find key features within each image, but the results don't compare to what the neural network models are learning. The LB score is only 0.63.

Basically, I tried to implement http://rajmadhan.myweb.usf.edu/sea_turtle.pdf

My notebook is @ https://github.com/daniel-bru/Turtle-SIFT_FacialRecognition/blob/main/Turtle_Facial_Recognition_SIFT_Similarity_.ipynb