Short hair and long hair transition
Now we have two Image Domain X (short hair) Y (long hair)
pretrained GAN image generated by a random vector, the vector representing each dimension of each feature of Image
If we know that certain (some) dimension represents the length of the hair, by adjusting this (these) dimensions, you can adjust the length of hair
So we need to decode Images
Generator (Decoder): pre-trained, can be converted into the random vector corresponding to the face
Encoder: Generator fixed parameters corresponding to the training vectors x Encoder generated by reconstructing Loss z x , z as only x to be able to produce the correct x by G
Discriminator: Discriminator when training Generator, can be used to initialize the parameters of Encoder
Said network are input to the short hair and long hair Images, which can be obtained, respectively Vectors
Vectors of the two Domain averaged, and subtracted, the distance can be obtained between the Domain z Long
Specific image X 0 , obtaining solutions coding En (x), plus × Z n- Long (as n-change parameter), to give z ', Gen (z') i.e. target image Y 0