def strongTransform(parameters, data=None, target=None):
assert ((data is not None) or (target is not None))
data, target = transformsgpu.mix(mask = parameters["Mix"], data = data, target = target)
data, target = transformsgpu.colorJitter(colorJitter = parameters["ColorJitter"], img_mean = torch.from_numpy(IMG_MEAN.copy()).cuda(), data = data, target = target)
data, target = transformsgpu.gaussian_blur(blur = parameters["GaussianBlur"], data = data, target = None)
data, target = transformsgpu.flip(flip = parameters["flip"], data = data, target = target)
return data, target
def weakTransform(parameters, data=None, target=None):
data, target = transformsgpu.flip(flip=parameters["flip"],
data=data,
target=target)
return data, target
def augmentationTransform(parameters,
data=None,
target=None,
probs=None,
jitter_vale=0.4,
min_sigma=0.2,
max_sigma=2.,
ignore_label=255):
"""
Args:
parameters: dictionary with the augmentation configuration
data: BxCxWxH input data to augment
target: BxWxH labels to augment
probs: BxWxH probability map to augment
jitter_vale: jitter augmentation value
min_sigma: min sigma value for blur
max_sigma: max sigma value for blur
ignore_label: value for ignore class
Returns:
augmented data, target, probs
"""
assert ((data is not None) or (target is not None))
if "Mix" in parameters:
data, target, probs = transformsgpu.mix(mask=parameters["Mix"],
data=data,
target=target,
probs=probs)
if "RandomScaleCrop" in parameters:
data, target, probs = transformsgpu.random_scale_crop(
scale=parameters["RandomScaleCrop"],
data=data,
target=target,
probs=probs,
ignore_label=ignore_label)
if "flip" in parameters:
data, target, probs = transformsgpu.flip(flip=parameters["flip"],
data=data,
target=target,
probs=probs)
if "ColorJitter" in parameters:
data, target, probs = transformsgpu.colorJitter(
colorJitter=parameters["ColorJitter"],
data=data,
target=target,
probs=probs,
s=jitter_vale)
if "GaussianBlur" in parameters:
data, target, probs = transformsgpu.gaussian_blur(
blur=parameters["GaussianBlur"],
data=data,
target=target,
probs=probs,
min_sigma=min_sigma,
max_sigma=max_sigma)
if "Grayscale" in parameters:
data, target, probs = transformsgpu.grayscale(
grayscale=parameters["Grayscale"],
data=data,
target=target,
probs=probs)
if "Solarize" in parameters:
data, target, probs = transformsgpu.solarize(
solarize=parameters["Solarize"],
data=data,
target=target,
probs=probs)
return data, target, probs