Coverage for hypergan/generators/dcgan_generator.py : 23%
Hot-keys on this page
r m x p toggle line displays
j k next/prev highlighted chunk
0 (zero) top of page
1 (one) first highlighted chunk
|
return []
gan = self.gan ops = self.ops config = self.config activation = ops.lookup(config.activation or 'lrelu')
print("[dcgan] NET IS", net)
net = ops.linear(net, 4*4*1024)
shape = ops.shape(net)
net = ops.reshape(net, [shape[0],4,4,1024])
net = activation(net) net = ops.deconv2d(net, 5, 5, 2, 2, 512) net = self.layer_filter(net) net = self.layer_regularizer(net) net = activation(net) net = ops.deconv2d(net, 5, 5, 2, 2, 256) net = self.layer_regularizer(net) net = activation(net) net = self.layer_filter(net) net = ops.deconv2d(net, 5, 5, 2, 2, 128) net = self.layer_regularizer(net) net = activation(net) net = self.layer_filter(net) net = ops.deconv2d(net, 5, 5, 2, 2, gan.channels()) net = self.layer_regularizer(net) net = ops.lookup('tanh')(net)
self.sample = net return self.sample |