How discriminator works on DCGAN?

Question

I studying about DCGAN, and I wonder something about it.

In Ian Goodfellow's natural GAN, discriminator Model outputs one scalar value what means the probability. But DCGAN's discriminator has designed with CNN architecture. I know that CNN's output is vector of class probabilities.

So how discriminator works on DCGAN? And what output of DCGAN's discriminator is?

I'm voting to close this question as off-topic because [it is about machine learning rather than software development](https://meta.stackoverflow.com/q/291009/1233251). You may find that there are many questions about generate adversarial networks on both [Cross Validated](//stats.stackexchange.com) and [DataScience.SE](//datascience.stackexchange.com). — E_net4, Dec 19 '17 at 14:06
You also seem to show a misconception: the output of a convolutional neural network is not always a vector a class probabilities. It is usually only so in a multi-class problem, when applying a softmax activation at the head of the network. — E_net4, Dec 19 '17 at 14:08

score 1 · Accepted Answer · answered Dec 21 '17 at 05:52

See Image Completion with Deep Learning in TensorFlow for a long answer.

In short: Suppose you make a CNN which has n filters of the size of its input and valid-padding. Then the output will be of shape n x 1 x 1. Then you can apply softmax to that shape and you have the probabilities in the channels.

You might also want to read 2.2.1. Convolutional Layers of my Masters thesis.

score 0 · Answer 2 · answered Dec 11 '18 at 00:52

Discriminator D takes a 3x64x64(for example) input image, processes it through a series of Conv2d, BatchNorm2d, and LeakyReLU layers, and outputs the final probability through a Sigmoid activation function.

Let's see an sample code to understand it's input and output.

class Discriminator(nn.Module):
def __init__(self, ngpu):
    super(Discriminator, self).__init__()
    self.ngpu = ngpu
    self.main = nn.Sequential(

        nn.Conv2d(nc, ndf, 4, 2, 1, bias=False),
        nn.LeakyReLU(0.2, inplace=True),

        nn.Conv2d(ndf, ndf*2, 4, 2, 1, bias=False),
        nn.BatchNorm2d(ndf*2),
        nn.LeakyReLU(0.2, inplace=True),

        nn.Conv2d(ndf*2, ndf*4, 4, 2, 1, bias=False),
        nn.BatchNorm2d(ndf*4),
        nn.LeakyReLU(0.2, inplace=True),

        nn.Conv2d(ndf*4, ndf*8, 4, 2, 1, bias=False ),
        nn.BatchNorm2d(ndf*8),
        nn.LeakyReLU(0.2, inplace=True),

        nn.Conv2d(ndf*8, 1, 4, 1, 0, bias=False),
        nn.Sigmoid()

    )

def forward(self, input):
    return self.main(input)

For more details, visit here

How discriminator works on DCGAN?

2 Answers2