def initialize_networks(self):
netD = networks.define_D2(self.opt) if self.opt.isTrain else None
if self.opt.isTrain and self.opt.which_iter_D2>0:
print_network(netD)
netD = util.load_network(netD, 'D2', self.opt.which_iter_D2 , self.opt)
return netD
def initialize_networks(self):
if 'cityscapes' in self.opt.dataset:
aspect_ratio = 2
else:
aspect_ratio = 1
generator = ProGANGenerator(max_dim=self.opt.max_dim, rgb=self.opt.rgb, aspect_ratio=aspect_ratio).cuda()
discriminator = ProGANDiscriminator(max_dim=self.opt.max_dim, rgb=self.opt.rgb, aspect_ratio=aspect_ratio).cuda()
print_network(generator)
print_network(discriminator)
return generator, discriminator
def initialize_networks(self, D_inputs=0):
if 'cityscapes' in self.opt.dataset:
aspect_ratio = 2
else:
aspect_ratio = 1
if not D_inputs:
D_inputs = self.opt.num_semantics
generator = ProGANGenerator(max_dim=self.opt.max_dim, rgb=self.opt.rgb, num_semantics=self.opt.num_semantics, T=self.opt.T, aspect_ratio=aspect_ratio).cuda()
discriminator = ProGANDiscriminator(max_dim=self.opt.max_dim, rgb=self.opt.rgb, num_semantics=D_inputs, aspect_ratio=aspect_ratio).cuda()
print_network(generator)
print_network(discriminator)
return generator, discriminator