Siamese Network to estimate similarity between RAW and GAN images (Python)
Budget: €30 – €250 EUR
We have a dataset of iris images that contain iris from RAW and GAN-generated images. The purpose of using the Siamese Network is to find the similarity rate between the left and right iris of the same person in both cases: 1- RAW image, 2- Fake image.
How? Training one network of Siamese with left iris images and training another network of Siamese with right iris images separately for RAW and GAN (Fake).
Estimated result? We should have a higher similarity rate for irises of RAW images (because they are symmetrical and similar), and a lower similarity rate for irises of Fake images (because they are generated by GAN and are not symmetrical/left and right iris of the same person are not similar in this case)
How? Training one network of Siamese with left iris images and training another network of Siamese with right iris images separately for RAW and GAN (Fake).
Estimated result? We should have a higher similarity rate for irises of RAW images (because they are symmetrical and similar), and a lower similarity rate for irises of Fake images (because they are generated by GAN and are not symmetrical/left and right iris of the same person are not similar in this case)