def _load_data(self, idx):
if self.inverse_crime:
#imgs, maps, masks = load_data_legacy(idx, self.data_file, self.gen_masks)
imgs, maps, masks = load_data(idx, self.data_file, self.gen_masks)
else:
imgs, maps, masks, ksp = load_data_ksp(idx, self.data_file,
self.gen_masks)
if self.scale_data:
## FIXME: batch mode
assert not self.scale_data, 'SEE FIXME'
sc = np.percentile(abs(imgs), 99, axis=(-1, -2))
imgs = imgs / sc
ksp = ksp / sc
if self.fully_sampled:
masks = np.ones(masks.shape)
if self.inverse_crime:
assert not self.noncart, 'FIXME: forward sim of NUFFT'
out = self._sim_data(imgs, maps, masks)
else:
out = self._sim_data(imgs, maps, masks, ksp)
if not self.noncart:
maps = fftmod(maps)
return imgs, maps, masks, out
def _sim_data(self, imgs, maps, masks, ksp=None):
# N, nc, nx, ny
if self.noncart:
assert ksp is not None, 'FIXME: NUFFT forward sim'
noise = np.random.randn(
*ksp.shape) + 1j * np.random.randn(*ksp.shape)
else:
noise = np.random.randn(
*maps.shape) + 1j * np.random.randn(*maps.shape)
if self.inverse_crime and ksp is None:
out = masks[:, None, :, :] * (fft2uc(imgs[:, None, :, :] * maps) +
1 / np.sqrt(2) * self.stdev * noise)
else:
if self.noncart:
out = ksp + 1 / np.sqrt(2) * self.stdev * noise
else:
out = masks[:, None, :, :] * (
ksp + 1 / np.sqrt(2) * self.stdev * noise)
if self.adjoint:
assert not self.noncart, 'FIXME: support NUFFT sim'
out = np.sum(np.conj(maps) * ifft2uc(out), axis=1).squeeze()
else:
if not self.noncart:
out = fftmod(out)
return out