discriminator output

[2018hello]

hi
Your work is great! Thanks a lot！
And I have a a question about discriminator output.
In the code, I see discriminator output a tensor ( batchsize1nn).
But I see a lot of gans output only a num ( batchsize1) in other code.
Why you use n*n? It can produce better results?

[knazeri]

@2018hello This class of discriminators is known as PatchGAN. The output of the discriminator is a nxn matrix with each index corresponding an image patch in the input. The receptive field of the network is 70x70 which means our discriminator evaluates 70x70 overlapping patches of the input image!

Philip Isola has a good explanation for this:

This is all mathematically equivalent to if we had manually chopped up the image into 70x70 overlapping patches, run a regular discriminator over each patch, and averaged the results.

[2018hello]

Thank you so much！ It help me a lot!