def test_random_distributions_2D(self):
### test whether all 4 possible mirrorings occur in approximately equal frquencies in 2D
batch_gen = BasicDataLoader((self.x_2D, self.y_2D), self.batch_size, number_of_threads_in_multithreaded=None)
batch_gen = SingleThreadedAugmenter(batch_gen, MirrorTransform((0, 1)))
counts = np.zeros(shape=(4,))
for b in range(self.num_batches):
batch = next(batch_gen)
for ix in range(self.batch_size):
if (batch['data'][ix, :, :, :] == self.cam_left).all():
counts[0] = counts[0] + 1
elif (batch['data'][ix, :, :, :] == self.cam_updown).all():
counts[1] = counts[1] + 1
elif (batch['data'][ix, :, :, :] == self.cam_updown_left).all():
counts[2] = counts[2] + 1
elif (batch['data'][ix, :, :, :] == self.cam).all():
counts[3] = counts[3] + 1
self.assertTrue([1 if (2200 < c < 2800) else 0 for c in counts] == [1]*4, "2D Images were not mirrored along "
"all axes with equal probability. "
"This may also indicate that "
"mirroring is not working")
def test_segmentations_2D(self):
### test whether segmentations are mirrored coherently with images
batch_gen = BasicDataLoader((self.x_2D, self.y_2D), self.batch_size, number_of_threads_in_multithreaded=None)
batch_gen = SingleThreadedAugmenter(batch_gen, MirrorTransform((0, 1)))
equivalent = True
for b in range(self.num_batches):
batch = next(batch_gen)
for ix in range(self.batch_size):
if (batch['data'][ix] != batch['seg'][ix]).all():
equivalent = False
self.assertTrue(equivalent, "2D images and seg were not mirrored in the same way (they should though because "
"seg needs to match the corresponding data")