def create_test_dataloaders(self):
if self.args.bayer:
transform_test = transforms.Compose([
transforms.ToTensor(),
preprocess_pg.PowerTransform(2.5),
preprocess_pg.Bayer([0.5, 0.5]),
])
else:
transform_test = transforms.Compose([
img_dataset.ToGrayscale(),
transforms.ToTensor(),
preprocess_pg.PowerTransform(2.5),
preprocess_pg.DiscreteIntensityScale([0.5, 1]),
])
testsets = [
(img_dataset.PlainImageFolder(root=denoising_data.set12_val_dir,
transform=transform_test,
cache=True), "Set12"),
(img_dataset.PlainImageFolder(
root=denoising_data.bsds500_val68_dir,
transform=transform_test,
cache=True), "val68"),
(img_dataset.PlainImageFolder(root=denoising_data.urban_val_dir,
transform=transform_test,
cache=True), "Urban100")
]
testloaders = [(torch.utils.data.DataLoader(testset,
batch_size=1,
shuffle=False,
num_workers=1), name)
for testset, name in testsets]
return testloaders
def create_test_dataloaders(self):
transform_test = transforms.Compose([
img_dataset.ToGrayscale(),
transforms.ToTensor(),
])
testsets = [
(img_dataset.PlainImageFolder(root=denoising_data.set12_val_dir,
transform=transform_test,
cache=True), "Set12"),
(img_dataset.PlainImageFolder(
root=denoising_data.bsds500_val68_dir,
transform=transform_test,
cache=True), "val68"),
(img_dataset.PlainImageFolder(root=denoising_data.urban_val_dir,
transform=transform_test,
cache=True), "Urban100")
]
testloaders = [(torch.utils.data.DataLoader(testset,
batch_size=1,
shuffle=False,
num_workers=1), name)
for testset, name in testsets]
return testloaders
def create_train_dataloaders(self, patchsize, batchsize, trainsetiters):
transform_train = transforms.Compose([
transforms.RandomCrop(patchsize),
preprocess.RandomOrientation90(),
transforms.RandomVerticalFlip(),
#img_dataset.ToGrayscale(),
transforms.ToTensor(),
])
self.batchsize = batchsize
train_folders = [
#denoising_data.bsds500_train_dir,
#denoising_data.bsds500_test_dir
#denoising_data.cmr_cine_train_dir
denoising_data.cmr_perf_train_dir
]
trainset = img_dataset.PlainImageFolder(root=train_folders,
transform=transform_train,
cache=True,
depth=2)
print('Input traing data has ', len(trainset))
trainset_multiple = [trainset] * trainsetiters
print(trainset_multiple)
trainset_used = torch.utils.data.ConcatDataset(trainset_multiple)
print('Total amount of images for training ', len(trainset_used))
# try to load all data
for n in tqdm(range(len(trainset_used))):
try:
img = trainset_used[n]
except:
print("Error in loading sample ", n)
trainloader = torch.utils.data.DataLoader(trainset_used,
batch_size=batchsize,
shuffle=True,
num_workers=20)
return trainloader
def create_train_dataloaders(self, patchsize, batchsize, trainsetiters):
if self.args.bayer:
transform_train = transforms.Compose([
transforms.RandomCrop(patchsize * 2),
preprocess.RandomOrientation90(),
transforms.RandomVerticalFlip(),
transforms.ToTensor(),
preprocess_pg.PowerTransform(1.25, 10),
preprocess_pg.Bayer([0.4, 0.7]),
])
else:
transform_train = transforms.Compose([
transforms.RandomCrop(patchsize),
preprocess.RandomOrientation90(),
transforms.RandomVerticalFlip(),
img_dataset.ToGrayscale(),
transforms.ToTensor(),
preprocess_pg.PowerTransform(1.25, 10),
preprocess_pg.ContinuousIntensityScale([0.25, 1]),
])
self.batchsize = batchsize
train_folders = [
denoising_data.bsds500_train_dir, denoising_data.bsds500_test_dir,
denoising_data.div2k_train_dir, denoising_data.waterloo_train_dir
]
trainset = img_dataset.PlainImageFolder(root=train_folders,
transform=transform_train,
cache=False)
trainset = torch.utils.data.ConcatDataset([trainset] * trainsetiters)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=batchsize,
shuffle=True,
num_workers=20)
return trainloader
def create_train_dataloaders(self, patchsize, batchsize, trainsetiters):
transform_train = transforms.Compose([
transforms.RandomCrop(patchsize),
preprocess.RandomOrientation90(),
transforms.RandomVerticalFlip(),
img_dataset.ToGrayscale(),
transforms.ToTensor(),
])
self.batchsize = batchsize
train_folders = [
denoising_data.bsds500_train_dir, denoising_data.bsds500_test_dir
]
trainset = img_dataset.PlainImageFolder(root=train_folders,
transform=transform_train,
cache=True)
trainset = torch.utils.data.ConcatDataset([trainset] * trainsetiters)
trainloader = torch.utils.data.DataLoader(trainset,
batch_size=batchsize,
shuffle=True,
num_workers=20)
return trainloader
