def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(
subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# now determine how much quality was reduced
applied_params = transformed_subject.applied_transforms[-1][1]
time = applied_params['times']['img']
degrees = np.sum(np.absolute(applied_params['degrees']['img']))
translation = np.sum(
np.absolute(applied_params['translation']['img']))
# motion in the middle of the acquisition process produces the most noticeable artifact
quality_reduction = clamp(degrees + translation, 0, 10) * min(
time, 1.0 - time)
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
if degrees + translation > 1: # it definitely has motion now
transformed_subject['info'][ghosting_motion_index] = 1
return transformed_subject
def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(
subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# now determine how much quality was reduced
applied_params = transformed_subject.applied_transforms[-1][1]
gamma = applied_params['gamma']['img'][0]
quality_reduction = 10 * abs(1.0 - gamma)
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
return transformed_subject
def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# now determine how much quality was reduced
applied_params = transformed_subject.applied_transforms[-1][1]
intensity = applied_params['intensity']['img']
num_ghosts = applied_params['num_ghosts']['img']
quality_reduction = 8 * intensity * math.log10(num_ghosts)
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
# it definitely has ghosting now
transformed_subject['info'][ghosting_motion_index] = 1
return transformed_subject
def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(
subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# now determine how much quality was reduced
applied_params = transformed_subject.applied_transforms[-1][1]
intensity = applied_params['intensity']['img']
# spikes_positions = applied_params['spikes_positions']['img']
quality_reduction = 0 + 2 * intensity
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
# it definitely has bias field now
transformed_subject['info'][inhomogeneity_index] = 1
return transformed_subject
def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(
subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# now determine how much quality was reduced
# applied_params = transformed_subject.applied_transforms[-1][1]
# coefficients = applied_params['coefficients']['img']
# quality_reduction = 2 + np.linalg.norm(np.asarray(coefficients))
quality_reduction = 4 # it is hard to assess impact on image quality
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
# it definitely has bias field now
transformed_subject['info'][inhomogeneity_index] = 1
return transformed_subject
def apply_transform(self, subject: torchio.Subject) -> torchio.Subject:
original_quality = subject['info'][0]
if original_quality < 6 and len(
subject.applied_transforms) == 1: # low quality image
return subject
else: # high-quality image, corrupt it
transformed_subject = super().apply_transform(subject)
# make sure we don't have negative intensities after adding noise
transformed_subject.img.set_data(
torch.clamp(transformed_subject.img.data, min=0.0, max=1.0))
# now determine how much quality was reduced
applied_params = transformed_subject.applied_transforms[-1][1]
std = applied_params['std']['img']
quality_reduction = 40 * std
# update the ground truth information
new_quality = original_quality - quality_reduction
transformed_subject['info'][0] = clamp(new_quality, 0, 10)
return transformed_subject