After reading this post, I sort of understand how the network transforms images; however, I cannot get how it actually LEARNS which orientation is helpful f
