def imagenormalize(data, divisor=None):
"""kspace generated by normalizing image space"""
#getting image from masked data
image = transforms.ifft2(data)
#normalizing the image
nimage, divisor = normalize(image, divisor)
#getting kspace data from normalized image
data = transforms.ifftshift(image, dim=(-3, -2))
data = torch.fft(data, 2)
data = transforms.fftshift(data, dim=(-3, -2))
return data, divisor
def mnormalize(masked_kspace):
#getting image from masked data
image = transforms.ifft2(masked_kspace)
#normalizing the image
nimage, mean, std = transforms.normalize_instance(image, eps=1e-11)
#getting kspace data from normalized image
maksed_kspace_fni = transforms.ifftshift(nimage, dim=(-3, -2))
maksed_kspace_fni = torch.fft(maksed_kspace_fni, 2)
maksed_kspace_fni = transforms.fftshift(maksed_kspace_fni, dim=(-3, -2))
maksed_kspace_fni, mean, std = transforms.normalize_instance(masked_kspace,
eps=1e-11)
return maksed_kspace_fni, mean, std
def onormalize(original_kspace, mean, std, eps=1e-11):
#getting image from masked data
image = transforms.ifft2(original_kspace)
#normalizing the image
nimage = transforms.normalize(image, mean, std, eps=1e-11)
#getting kspace data from normalized image
original_kspace_fni = transforms.ifftshift(nimage, dim=(-3, -2))
original_kspace_fni = torch.fft(original_kspace_fni, 2)
original_kspace_fni = transforms.fftshift(original_kspace_fni,
dim=(-3, -2))
original_kspace_fni = transforms.normalize(original_kspace,
mean,
std,
eps=1e-11)
return original_kspace_fni
def nkspacetoimage(args, kspace_fni, mean, std, eps=1e-11):
#nkspace to image
assert kspace_fni.size(-1) == 2
image = transforms.ifftshift(kspace_fni, dim=(-3, -2))
image = torch.ifft(image, 2)
image = transforms.fftshift(image, dim=(-3, -2))
#denormalizing the nimage
image = (image * std) + mean
image = image[0]
image = transforms.complex_center_crop(image,
(args.resolution, args.resolution))
# Absolute value
image = transforms.complex_abs(image)
# Normalize input
image, mean, std = transforms.normalize_instance(image, eps=1e-11)
image = image.clamp(-6, 6)
return image
class C3Convert:
def __init__(self, shp=(320, 320)):
self.c3m = transforms.c3_torch(shp) # ksp.shape[-2:]
def apply(self, ksp):
# expect (bat, w, h, 2)
return ksp * self.c3m
# torch fft requires (bat, w,h, 2)
ifft_c3 = lambda kspc3: torch.ifft(
transforms.ifftshift(kspc3, dim=(-3, -2)), 2, normalized=True)
fft_c3 = lambda im: transforms.fftshift(torch.fft(im, 2, normalized=True),
dim=(-3, -2))
class NoTransform:
def __init__(self):
pass
def __call__(self, kspace, target, attrs, fname, slice):
return kspace
class BasicMaskingTransform:
def __init__(self,
mask_func,
resolution,
which_challenge,
def test_fftshift(shape):
input = np.arange(np.product(shape)).reshape(shape)
out_torch = transforms.fftshift(torch.from_numpy(input)).numpy()
out_numpy = np.fft.fftshift(input)
assert np.allclose(out_torch, out_numpy)