Is your feature request related to a problem? Please describe.
For data augmentation, you might want to flip your images. You might want to flip it around any of the dimensions, either all at the same time or with some combination. The current code, however, only allows for flipping across all or none of the dimensions.
The current solution is therefore to chain multiple transforms together, but I feel a single transform should be able to give this functionality.
Current workaround:
RandFlipd(["image", "label"], prob=0.5, spatial_axis=0),
RandFlipd(["image", "label"], prob=0.5, spatial_axis=1),
Describe the solution you'd like
def randomize(data):
if isinstance(self.spatial_axis, int):
self.sampled_spatial_axis = self.spatial_axis
else:
sample = np.random.randint(2, size=len(self.sampled_spatial_axis))
self.sampled_spatial_axis = np.array(self.sampled_spatial_axis)[sample].tolist()
Is your feature request related to a problem? Please describe.
For data augmentation, you might want to flip your images. You might want to flip it around any of the dimensions, either all at the same time or with some combination. The current code, however, only allows for flipping across all or none of the dimensions.
The current solution is therefore to chain multiple transforms together, but I feel a single transform should be able to give this functionality.
Current workaround:
Describe the solution you'd like