def johnson_alahi_li_2016_content_transform(
edge_size: int = 256, multiple: int = 16, impl_params: bool = True,
) -> ComposedTransform:
class TopLeftCropToMultiple(Transform):
def __init__(self, multiple: int):
super().__init__()
self.multiple = multiple
def calculate_size(self, image: torch.Tensor) -> Tuple[int, int]:
old_height, old_width = extract_image_size(image)
new_height = old_height - old_height % self.multiple
new_width = old_width - old_width % self.multiple
return new_height, new_width
def forward(self, image: torch.tensor) -> torch.Tensor:
size = self.calculate_size(image)
return top_left_crop(image, size)
class OptionalGrayscaleToFakegrayscale(Transform):
def forward(self, input_image: torch.Tensor) -> torch.Tensor:
is_grayscale = extract_num_channels(input_image) == 1
if is_grayscale:
return grayscale_to_fakegrayscale(input_image)
else:
return input_image
transforms = [
TopLeftCropToMultiple(multiple),
Resize((edge_size, edge_size)),
OptionalGrayscaleToFakegrayscale(),
]
if impl_params:
transforms.append(CaffePreprocessing())
return ComposedTransform(*transforms)
def li_wand_2016_preprocessor() -> CaffePreprocessing:
return CaffePreprocessing()
def ulyanov_et_al_2016_preprocessor() -> CaffePreprocessing:
return CaffePreprocessing()
def johnson_alahi_li_2016_preprocessor() -> CaffePreprocessing:
return CaffePreprocessing()
def gatys_et_al_2017_preprocessor() -> CaffePreprocessing:
return CaffePreprocessing()
def gatys_ecker_bethge_2015_preprocessor() -> nn.Module:
return CaffePreprocessing()