import pickle
import numpy as np
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from libs.Matrix import MulLayer
import torchvision.utils as vutils
import torch.backends.cudnn as cudnn
from libs.models import encoder3, encoder4
from libs.models import decoder3, decoder4
os.environ[
"CUDA_VISIBLE_DEVICES"] = "0" # USED ONLY IF OTHER GPUS ARE BEING USED
if True:
style_dataset = Dataset('Database/WikiArt/train/', 256, 256, test=True)
style_loader_ = torch.utils.data.DataLoader(dataset=style_dataset,
batch_size=64,
shuffle=False,
num_workers=4,
drop_last=True)
style_loader = iter(style_loader_)
# styleV = torch.Tensor(64,3,224,224).cuda()
matrix = MulLayer('r41')
vgg = encoder4()
vgg.load_state_dict(torch.load('models/vgg_r41.pth'))
matrix.load_state_dict(torch.load('models/r41.pth'))
vgg.cuda()
matrix.cuda()
features = []
# help='which features to transfer, either r31 or r41')
parser.add_argument("--layer",
default="r41",
help='which features to transfer, either r31 or r41')
################# PREPARATIONS #################
opt = parser.parse_args()
opt.cuda = torch.cuda.is_available()
print_options(opt)
os.makedirs(opt.outf, exist_ok=True)
cudnn.benchmark = True
################# DATA #################
content_dataset = Dataset(opt.contentPath,
opt.loadSize,
opt.fineSize,
test=True)
content_loader = torch.utils.data.DataLoader(dataset=content_dataset,
batch_size=opt.batchSize,
shuffle=False,
num_workers=1)
style_dataset = Dataset(opt.stylePath,
opt.loadSize,
opt.fineSize,
test=True)
style_loader = torch.utils.data.DataLoader(dataset=style_dataset,
batch_size=opt.batchSize,
shuffle=False,
num_workers=1)
################# MODEL #################
torch.cuda.FloatTensor(self.features[idx]).shape, sF.shape,
Fm.shape)
else:
idx = np.random.randint(0, self.size, b)
sF = torch.cuda.FloatTensor(self.features[idx])
Fm = torch.cuda.FloatTensor(self.means[idx])
cF = self.vgg(x)
feature, transmatrix = self.matrix(cF, sF, Fm)
transfer = self.dec(feature)
return transfer.clamp(0, 1)
if __name__ == "__main__":
batch_size = 8
content_dataset = Dataset('Database/COCO/2017/train2017/',
256,
256,
test=True)
content_loader = torch.utils.data.DataLoader(dataset=content_dataset,
batch_size=batch_size,
shuffle=False,
num_workers=1,
drop_last=True)
Stylenet = StyleAugmentation().cuda()
for it, (content, _) in enumerate(content_loader):
styled = Stylenet(content.cuda())
# vutils.save_image(styled[0].data,'Style.png',normalize=True,scale_each=True,nrow=1)
for n in range(styled.shape[0]):
Image = np.uint8(
content.permute(0, 2, 3, 1)[n].cpu().detach().numpy() * 255)
