def get_transforms(mode="train", target_size=128):
tranform_list = []
if mode == "train":
tranform_list = [# CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=target_size, order=1),
MirrorTransform(axes=(1,)),
SpatialTransform(patch_size=(target_size,target_size), random_crop=False,
patch_center_dist_from_border=target_size // 2,
do_elastic_deform=True, alpha=(0., 1000.), sigma=(40., 60.),
do_rotation=True, p_rot_per_sample=0.5,
angle_x=(-0.1, 0.1), angle_y=(0, 1e-8), angle_z=(0, 1e-8),
scale=(0.5, 1.9), p_scale_per_sample=0.5,
border_mode_data="nearest", border_mode_seg="nearest"),
]
elif mode == "val":
tranform_list = [CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=target_size, order=1),
]
elif mode == "test":
tranform_list = [CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=target_size, order=1),
]
tranform_list.append(NumpyToTensor())
return Compose(tranform_list)
def get_transforms(mode="train", target_size=128):
tranform_list = []
if mode == "train":
tranform_list = [ # CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=(target_size, target_size), order=1),
MirrorTransform(axes=(1, )),
]
elif mode == "val":
tranform_list = [ #CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=target_size, order=1),
MirrorTransform(axes=(1, )),
]
elif mode == "test":
tranform_list = [ #CenterCropTransform(crop_size=target_size),
ResizeTransform(target_size=target_size, order=1),
MirrorTransform(axes=(1, )),
]
tranform_list.append(NumpyToTensor())
return Compose(tranform_list)
dataset_test = ConditionalGanDataset(path_test_real, load_sample_cgan_test, ['.PNG', '.png'], ['.PNG', '.png'])
### Transforms applied to data
from batchgenerators.transforms import RandomCropTransform, Compose
from batchgenerators.transforms.spatial_transforms import ResizeTransform,SpatialTransform
transforms = Compose([
#SpatialTransform(patch_size=(1024, 1024), do_rotation=True, patch_center_dist_from_border=1024, border_mode_data='reflect',
# border_mode_seg='reflect', angle_x=(args.rot_angle, args.rot_angle), angle_y=(0, 0), angle_z=(0, 0),
# do_elastic_deform=False, order_data=1, order_seg=1)
ResizeTransform((int(args.resize_size), int(args.resize_size)), order=1),
RandomCropTransform((params.nested_get("image_size"), params.nested_get("image_size"))),
])
from delira.data_loading import BaseDataManager, SequentialSampler, RandomSampler
manager_test = BaseDataManager(dataset_test, params.nested_get("batch_size"),
transforms=transforms,
sampler_cls=SequentialSampler,
n_process_augmentation=1)
import warnings
warnings.simplefilter("ignore", UserWarning) # ignore UserWarnings raised by dependency code
warnings.simplefilter("ignore", FutureWarning) # ignore FutureWarnings raised by dependency code
path_val.append(os.path.join(root_path, dataset_name + '/Original50mm/'))
"""
dataset_train = ConditionalGanDataset(path_train, load_sample_cgan,
['.PNG', '.png'], ['.PNG', '.png'])
dataset_val = ConditionalGanDataset(path_val, load_sample_cgan,
['.PNG', '.png'], ['.PNG', '.png'])
### Transforms applied to data
from batchgenerators.transforms import RandomCropTransform, Compose
from batchgenerators.transforms.spatial_transforms import ResizeTransform
transforms = Compose([
ResizeTransform((1024, 1024), order=1),
RandomCropTransform(
(params.nested_get("image_size"), params.nested_get("image_size")
)), # Perform Random Crops of Size 256 x 256 pixels,
])
### Data manager
from delira.data_loading import BaseDataManager, SequentialSampler, RandomSampler
manager_train = BaseDataManager(dataset_train,
params.nested_get("batch_size"),
transforms=transforms,
sampler_cls=RandomSampler,
n_process_augmentation=4)
def get_transforms(mode="train",
n_channels=1,
target_size=128,
add_resize=False,
add_noise=False,
mask_type="",
batch_size=16,
rotate=True,
elastic_deform=True,
rnd_crop=False,
color_augment=True):
tranform_list = []
noise_list = []
if mode == "train":
tranform_list = [
FillupPadTransform(min_size=(n_channels, target_size + 5,
target_size + 5)),
ResizeTransform(target_size=(target_size + 1, target_size + 1),
order=1,
concatenate_list=True),
# RandomCropTransform(crop_size=(target_size + 5, target_size + 5)),
MirrorTransform(axes=(2, )),
ReshapeTransform(new_shape=(1, -1, "h", "w")),
SpatialTransform(patch_size=(target_size, target_size),
random_crop=rnd_crop,
patch_center_dist_from_border=target_size // 2,
do_elastic_deform=elastic_deform,
alpha=(0., 100.),
sigma=(10., 13.),
do_rotation=rotate,
angle_x=(-0.1, 0.1),
angle_y=(0, 1e-8),
angle_z=(0, 1e-8),
scale=(0.9, 1.2),
border_mode_data="nearest",
border_mode_seg="nearest"),
ReshapeTransform(new_shape=(batch_size, -1, "h", "w"))
]
if color_augment:
tranform_list += [ # BrightnessTransform(mu=0, sigma=0.2),
BrightnessMultiplicativeTransform(multiplier_range=(0.95, 1.1))
]
tranform_list += [
GaussianNoiseTransform(noise_variance=(0., 0.05)),
ClipValueRange(min=-1.5, max=1.5),
]
noise_list = []
if mask_type == "gaussian":
noise_list += [GaussianNoiseTransform(noise_variance=(0., 0.2))]
elif mode == "val":
tranform_list = [
FillupPadTransform(min_size=(n_channels, target_size + 5,
target_size + 5)),
ResizeTransform(target_size=(target_size + 1, target_size + 1),
order=1,
concatenate_list=True),
CenterCropTransform(crop_size=(target_size, target_size)),
ClipValueRange(min=-1.5, max=1.5),
# BrightnessTransform(mu=0, sigma=0.2),
# BrightnessMultiplicativeTransform(multiplier_range=(0.95, 1.1)),
CopyTransform({"data": "data_clean"}, copy=True)
]
noise_list += []
if add_noise:
tranform_list = tranform_list + noise_list
tranform_list.append(NumpyToTensor())
return Compose(tranform_list)