Training procedure

[dcnocomment]

Why there are 2 training files(train_128 and train_x2). I notice that the second layer receives first layer output:

Data-set pair is line drawing & colored img 1st layer : 128128 input line drawing gray-scale 1ch & random picked src px 3ch output 3ch img 2nd layer : 512512 input line drawing gray-scale 1ch & 1st layer output 3ch output 3ch img

but why they are trained separately? and what is the meaning of the 2 layers?

[taizan]

the 1st layer is deciding basic color and 2nd layer is upscaling it for better result. training 2 network at same time is very slow and unstable. ( recently some connected learning method is proposed for high-resolution images )