def __init__(self, patch_size=8, overlap=4):
"""
Args:
patch_size: size of patches when cutting up the image into overlapping patches
overlap: how much to overlap patches
"""
if not _TORCHVISION_AVAILABLE:
raise ModuleNotFoundError( # pragma: no-cover
'You want to use `transforms` from `torchvision` which is not installed yet.'
)
self.patch_size = patch_size
self.overlap = overlap
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply(
[transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([RandomTranslateWithReflect(4)],
p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
self.transforms = transforms.Compose([
img_jitter,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize,
Patchify(patch_size=patch_size, overlap_size=overlap),
])
def __init__(self, patch_size=8):
# flipping image along vertical axis
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
# image augmentation functions
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply(
[transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([RandomTranslateWithReflect(4)],
p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
# main transform for self-supervised training
self.train_transform = transforms.Compose([
img_jitter,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize,
Patchify(patch_size=patch_size, overlap_size=patch_size // 2),
])
# transform for testing
self.test_transform = transforms.Compose([
transforms.ToTensor(),
normalize,
Patchify(patch_size=patch_size, overlap_size=patch_size // 2),
])
def __init__(self, patch_size=8, overlap=4):
"""
Transforms used for CPC:
Args:
patch_size: size of patches when cutting up the image into overlapping patches
overlap: how much to overlap patches
Transforms::
random_flip
img_jitter
col_jitter
rnd_gray
transforms.ToTensor()
normalize
Patchify(patch_size=patch_size, overlap_size=patch_size // 2)
Example::
# in a regular dataset
CIFAR10(..., transforms=CPCTrainTransformsCIFAR10())
# in a DataModule
module = CIFAR10DataModule(PATH)
train_loader = module.train_dataloader(batch_size=32, transforms=CPCTrainTransformsCIFAR10())
"""
if not _TORCHVISION_AVAILABLE:
raise ImportError(
'You want to use `transforms` from `torchvision` which is not installed yet.'
)
self.patch_size = patch_size
self.overlap = overlap
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply(
[transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([RandomTranslateWithReflect(4)],
p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
self.transforms = transforms.Compose([
img_jitter,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize,
Patchify(patch_size=patch_size, overlap_size=overlap),
])
def __init__(self):
"""
Transforms applied to AMDIM
Transforms::
img_jitter,
col_jitter,
rnd_gray,
transforms.ToTensor(),
normalize
Example::
x = torch.rand(5, 3, 32, 32)
transform = AMDIMTrainTransformsCIFAR10()
(view1, view2) = transform(x)
"""
if not _TORCHVISION_AVAILABLE:
raise ImportError(
'You want to use `transforms` from `torchvision` which is not installed yet.'
)
# flipping image along vertical axis
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
# image augmentation functions
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply(
[transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([RandomTranslateWithReflect(4)],
p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
self.transforms = transforms.Compose([
img_jitter, col_jitter, rnd_gray,
transforms.ToTensor(), normalize
])
def __init__(self):
if not _TORCHVISION_AVAILABLE:
raise ModuleNotFoundError( # pragma: no-cover
'You want to use `transforms` from `torchvision` which is not installed yet.'
)
# flipping image along vertical axis
self.flip_lr = transforms.RandomHorizontalFlip(p=0.5)
# image augmentation functions
normalize = transforms.Normalize(
mean=[x / 255.0 for x in [125.3, 123.0, 113.9]],
std=[x / 255.0 for x in [63.0, 62.1, 66.7]])
col_jitter = transforms.RandomApply(
[transforms.ColorJitter(0.4, 0.4, 0.4, 0.2)], p=0.8)
img_jitter = transforms.RandomApply([RandomTranslateWithReflect(4)],
p=0.8)
rnd_gray = transforms.RandomGrayscale(p=0.25)
self.transforms = transforms.Compose([
img_jitter, col_jitter, rnd_gray,
transforms.ToTensor(), normalize
])