Hi Mr. Liu,
I'm interested in your work and tried your code on svhn2mnist task. Here are several generated images and I want to ask why the a2b and b2a from decode_b(ca, sb) decode_a(cb, sa) are bad?
But decode_b(ca, noise) decode_a(cb, noise) are good but not relavant to the original class label?
The first ten labels of svhn images are 8, 9, 1, 4, 3, 5, 3, 8, 5, 3 ....
Hi Mr. Liu, I'm interested in your work and tried your code on svhn2mnist task. Here are several generated images and I want to ask why the a2b and b2a from decode_b(ca, sb) decode_a(cb, sa) are bad? But decode_b(ca, noise) decode_a(cb, noise) are good but not relavant to the original class label?
The first ten labels of svhn images are 8, 9, 1, 4, 3, 5, 3, 8, 5, 3 ....