default=1,
help='hyperparameter to blend wct feature and content feature')
parser.add_argument('--gpu',
type=int,
default=0,
help="which gpu to run on. default is 0")
args = parser.parse_args()
try:
os.makedirs(args.outf)
except OSError:
pass
# Data loading code
dataset = Dataset(args.contentPath, args.stylePath, args.fineSize)
loader = torch.utils.data.DataLoader(dataset=dataset,
batch_size=1,
shuffle=False)
wct = WCT(args)
def styleTransfer(contentImg, styleImg, imname, csF):
sF5 = wct.e5(styleImg)
cF5 = wct.e5(contentImg)
sF5 = sF5.data.cpu().squeeze(0)
cF5 = cF5.data.cpu().squeeze(0)
csF5 = wct.transform(cF5, sF5, csF, args.alpha)
Im5 = wct.d5(csF5)
def default_loader(path):
return Image.open(path).convert('RGB')
if __name__ == "__main__":
model = ResNet_autoencoder(Bottleneck, DeconvBottleneck, [3, 4, 6, 3],
3).cuda()
model.load_state_dict(torch.load(PATH))
model.eval()
# Data loading code
dataset = Dataset(
"./content",
"./style",
)
loader = torch.utils.data.DataLoader(dataset=dataset,
batch_size=1,
shuffle=False)
alpha_list = [0.4, 0.6, 0.7, 0.8, 1.0]
for alpha in alpha_list:
for i, (contentImg, styleImg, imname) in enumerate(loader):
imname = imname[0]
print('Transferring ' + imname)
contentImg = contentImg.cuda()
styleImg = styleImg.cuda()
cImg = Variable(contentImg, volatile=True)
sImg = Variable(styleImg, volatile=True)
c, s = model.encoder(cImg), model.encoder(sImg)
type=int,
default=-1,
help="Multi = -1, else for specific level use level = {1,2,3,4,5}")
parser.add_argument('--do_patches', type=bool, default=False)
parser.add_argument('--kernel_size', type=int, default=512)
parser.add_argument('--stride', type=int, default=512)
args = parser.parse_args()
try:
os.makedirs(args.outf)
except OSError:
pass
# Data loading code
dataset = Dataset(args.contentPath, args.stylePath, args.fineSize,
args.do_patches, args.kernel_size)
loader = torch.utils.data.DataLoader(dataset=dataset,
batch_size=1,
shuffle=False)
wct = WCT(args)
def styleTransfer(contentImg,
styleImg,
imname,
csF,
level=-1,
save_path=args.outf):
# print("level= ",level)
if (level == -1):
# parser.add_argument('--d3', default='../KD/Bin/models/small10x_decoder/3SD_10x_E29S5000.pth')
# parser.add_argument('--d2', default='../KD/Bin/models/my_decoder/2BD_E30S0.pth')
# parser.add_argument('--d1', default='../KD/Bin/models/my_decoder/1BD_E30S0.pth')
args = parser.parse_args()
try:
os.makedirs(args.outf)
except OSError:
pass
# Data loading code
contentPath = args.UHD_contentPath if args.UHD else args.contentPath
stylePath = args.UHD_stylePath if args.UHD else args.stylePath
dataset = Dataset(contentPath, stylePath, args.texturePath, args.fineSize,
args.picked_content_mark, args.picked_style_mark,
args.synthesis)
loader = torch.utils.data.DataLoader(dataset=dataset,
batch_size=1,
shuffle=False)
logprint(args.log_mark)
log = open("samples/log_%s_%s.txt" % (args.mode, args.log_mark), "w+")
logprint(args._get_kwargs(), log)
wct = WCT(args)
@torch.no_grad()
def styleTransfer(encoder, decoder, contentImg, styleImg, csF):
# sF = encoder.forward_aux(styleImg)[-1]; torch.cuda.empty_cache()
