load_options.dataroot, "train",
torchvision.transforms.Compose([
torchvision.transforms.Resize(opt.sizex, Image.NEAREST),
torchvision.transforms.CenterCrop(opt.sizex),
torchvision.transforms.ToTensor(),
]))
if opt.dataset == 'flowers':
dataset = datasets.FlowersDataset(
load_options.dataroot, "train",
torchvision.transforms.Compose([
torchvision.transforms.Resize(opt.sizex, Image.NEAREST),
torchvision.transforms.CenterCrop(opt.sizex),
torchvision.transforms.ToTensor(),
]))
if opt.dataset == 'cmnist':
dataset = datasets.CMNISTDataset(dataPath=load_options.dataroot,
sets='train')
loader = torch.utils.data.DataLoader(dataset,
batch_size=load_options.batch_size,
shuffle=True)
xData, mData = next(iter(loader))
xData = xData.to(device)
mData = mData.to(device)
## Use the same z for all images in batch: ##
# z = torch.randn(1, opt.nMasks, opt.nz, 1, 1).repeat(batch_size, 1, 1, 1, 1).to(device)
## or use different z: ##
z = torch.randn(load_options.batch_size, opt.nMasks, opt.nz, 1, 1).to(device)
with torch.no_grad():
transforms.ToTensor(),
]))
testset = datasets.FlowersDataset(opt.dataroot,
"test",
transforms.Compose([transforms.Resize(opt.sizex, Image.NEAREST),
transforms.CenterCrop(opt.sizex),
transforms.ToTensor(),
]))
valset = datasets.FlowersDataset(opt.dataroot,
"val",
transforms.Compose([transforms.Resize(opt.sizex, Image.NEAREST),
transforms.CenterCrop(opt.sizex),
transforms.ToTensor(),
]))
if opt.dataset == 'cmnist':
trainset = datasets.CMNISTDataset(dataPath=opt.dataroot,
sets='train')
testset = datasets.CMNISTDataset(dataPath=opt.dataroot,
sets='test')
valset = datasets.CMNISTDataset(dataPath=opt.dataroot,
sets='val')
trainloader = torch.utils.data.DataLoader(trainset, batch_size=opt.batchSize, shuffle=True, num_workers=4, drop_last=True)
def weights_init_ortho(m):
if isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d) or isinstance(m, nn.Linear):
nn.init.orthogonal_(m.weight, opt.initOrthoGain)
netEncM = models._netEncM(sizex=opt.sizex, nIn=opt.nx, nMasks=opt.nMasks, nRes=opt.nResM, nf=opt.nfM, temperature=opt.temperature).to(device)
netGenX = models._netGenX(sizex=opt.sizex, nOut=opt.nx, nc=opt.nz, nf=opt.nfX, nMasks=opt.nMasks, selfAtt=opt.useSelfAttG).to(device)
netDX = models._resDiscriminator128(nIn=opt.nx, nf=opt.nfD, selfAtt=opt.useSelfAttD).to(device)
transforms.Resize(opt.sizex, Image.NEAREST),
transforms.CenterCrop(opt.sizex),
transforms.ToTensor(),
]),
)
valset = datasets.FlowersDataset(
opt.dataroot,
"val",
transforms.Compose([
transforms.Resize(opt.sizex, Image.NEAREST),
transforms.CenterCrop(opt.sizex),
transforms.ToTensor(),
]),
)
if opt.dataset == "cmnist":
trainset = datasets.CMNISTDataset(dataPath=opt.dataroot, sets="train")
testset = datasets.CMNISTDataset(dataPath=opt.dataroot, sets="test")
valset = datasets.CMNISTDataset(dataPath=opt.dataroot, sets="val")
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=opt.batchSize,
shuffle=True,
num_workers=4,
drop_last=True)
def weights_init_ortho(m):
if (isinstance(m, nn.Conv2d) or isinstance(m, nn.ConvTranspose2d)
or isinstance(m, nn.Linear)):
nn.init.orthogonal_(m.weight, opt.initOrthoGain)