I am trying to create an image completion such as the one presented in ImageCompletion Using DCGAN, being implemented in Matlab using the CNN toolbox (in order to not use MatConvNet). I am quite sure that with regression layer it is possible to generate regions of interest (RoIs), but, what i am doing, is to search for missing parts of an object, for example, a cat without face, and put a face of a cat in it.
I know that semantic segmentation give an image as output, but further code is needed to generate the problem that i'm facing.
So my question is, is there a possibility of get an output image, locating the missing part of the object?
Because another problem, is the arrange of the data entered to the network for training.
Thanks in advance for the help given.