def test_background_usage(self):
img = prepare_image((80, 80))
gt = np.zeros((80, 80), dtype=int)
background = np.ones((80, 80))
background = background * 0.5
draw_cell(img, (10, 10), 6)
draw_disc(gt, (10, 10), 6, 1)
draw_cell(img, (30, 15), 8)
draw_disc(gt, (30, 15), 6, 1)
# background with cell looking noise
draw_cell(img, (30, 40), 9)
draw_cell(background, (30, 40), 9)
cellstar = Segmentation(9, 14)
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
# background noise is returned as cell
self.assertEqual(3, segmentation.max())
self.assertEqual(3, len(snakes))
cellstar = Segmentation(9, 14)
cellstar.set_frame(img)
cellstar.set_background(background)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(2, segmentation.max())
self.assertEqual(2, len(snakes))
def test_mask_usage(self):
img = prepare_image((80, 80))
gt = np.zeros((80, 80), dtype=int)
draw_very_weak_cell(img, (10, 10), 6)
draw_disc(gt, (10, 10), 6, 1)
draw_very_weak_cell(img, (30, 15), 8)
draw_disc(gt, (30, 15), 6, 1)
# very bright points
img[50:70, 50:70] = 1
img[50:70, 10:30] = 0
cellstar = Segmentation(9, 10)
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(0, segmentation.max())
self.assertEqual(0, len(snakes))
ignore_mask = (img == 1) | (img == 0)
cellstar = Segmentation(9, 10)
cellstar.set_frame(img)
cellstar.set_mask(ignore_mask)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(2, segmentation.max())
self.assertEqual(2, len(snakes))
def test_no_objects(self):
img = prepare_image((50, 50))
cellstar = Segmentation(9, 10)
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(0, segmentation.max())
self.assertEqual(0, len(snakes))
def test_multiple_objects(self):
img = prepare_image((50, 50))
draw_cell(img, (10, 10), 6)
draw_cell(img, (30, 15), 8)
cellstar = Segmentation(9, 10)
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(2, segmentation.max())
self.assertEqual(2, len(snakes))
def test_trained_parameters(image, star_params, ranking_params, precision, avg_cell_diameter, output_name=None):
seg = Segmentation(segmentation_precision=precision, avg_cell_diameter=avg_cell_diameter)
for k, v in star_params.iteritems():
seg.parameters["segmentation"]["stars"][k] = v
for k, v in ranking_params.iteritems():
seg.parameters["segmentation"]["ranking"][k] = v
seg.set_frame(image)
seg.run_segmentation()
if output_name is None:
image_show(seg.images.segmentation, 1)
else:
image_save(seg.images.segmentation, output_name)
def test_rank_fitting(self):
img = prepare_image((80, 80))
gt = np.zeros((80, 80), dtype=int)
draw_weak_cell(img, (60, 20), 12)
draw_disc(gt, (60, 20), 10, 1)
draw_weak_cell(img, (40, 25), 8)
draw_disc(gt, (40, 25), 6, 2)
draw_weak_cell(img, (30, 40), 9)
draw_disc(gt, (30, 40), 7, 3)
img = finish_image(img)
cellstar = Segmentation(11, 20)
# break parameters
encoded = params.pf_rank_parameters_encode(cellstar.parameters)
one_params = [0.1 for k in encoded]
one_decoded = params.pf_rank_parameters_decode(one_params)
broken_params = PFRankSnake.merge_rank_parameters(
cellstar.parameters, one_decoded)
cellstar.parameters = broken_params
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
# fail miserably (best snakes are nowhere close)
best3 = get_best_mask(segmentation, 3)
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(segmentation_quality, 0.01)
new_params, _ = run_rank_pf(img, None, None, gt, cellstar.parameters)
cellstar = Segmentation(11, 20)
cellstar.parameters = new_params
cellstar.set_frame(img)
segmentation2, snakes2 = cellstar.run_segmentation()
self.assertLessEqual(3, segmentation2.max())
self.assertLessEqual(3, len(snakes2))
# find best 3 objects
best3 = get_best_mask(segmentation2, 3)
snake3 = snakes2[:3]
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(
0.3, segmentation_quality) # there was no contour fitting
object_diffs = calculate_diffs_per_object(best3, gt)
self.assertLessEqual(0.3, min(object_diffs))
def test_contour_fitting(self):
process.SEARCH_LENGTH_NORMAL = 20
img = prepare_image((80, 80))
gt = np.zeros((80, 80), dtype=int)
draw_weak_cell(img, (60, 20), 12)
draw_disc(gt, (60, 20), 14, 1)
draw_weak_cell(img, (40, 25), 8)
draw_disc(gt, (40, 25), 8, 2)
draw_weak_cell(img, (30, 40), 9)
draw_disc(gt, (30, 40), 9, 3)
img = finish_image(img)
cellstar = Segmentation(11, 20)
# break parameters
weights = cellstar.parameters["segmentation"]["stars"]["sizeWeight"]
encoded = params.pf_parameters_encode(cellstar.parameters)
one_params = [0.5 for k in encoded]
one_params[3] = 0.001
one_decoded = params.pf_parameters_decode(one_params, weights)
broken_params = PFSnake.merge_parameters(cellstar.parameters,
one_decoded)
cellstar.parameters = broken_params
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
# fail miserably (not all snakes are found)
self.assertGreater(3, segmentation.max())
self.assertGreater(3, len(snakes))
new_params, _ = run_pf(img, None, None, gt, cellstar.parameters, 11,
20)
cellstar = Segmentation(11, 20)
cellstar.parameters = new_params
cellstar.set_frame(img)
segmentation2, snakes2 = cellstar.run_segmentation()
self.assertLessEqual(3, segmentation2.max())
self.assertLessEqual(3, len(snakes2))
# find best 3 objects
best3 = get_best_mask(segmentation2, 3)
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(0.7, segmentation_quality)
def test_rank_fitting(self):
img = prepare_image((80, 80))
gt = np.zeros((80, 80), dtype=int)
draw_weak_cell(img, (60, 20), 12)
draw_disc(gt, (60, 20), 10, 1)
draw_weak_cell(img, (40, 25), 8)
draw_disc(gt, (40, 25), 6, 2)
draw_weak_cell(img, (30, 40), 9)
draw_disc(gt, (30, 40), 7, 3)
img = finish_image(img)
cellstar = Segmentation(11, 20)
# break parameters
encoded = params.pf_rank_parameters_encode(cellstar.parameters)
one_params = [0.1 for k in encoded]
one_decoded = params.pf_rank_parameters_decode(one_params)
broken_params = PFRankSnake.merge_rank_parameters(cellstar.parameters, one_decoded)
cellstar.parameters = broken_params
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
# fail miserably (best snakes are nowhere close)
best3 = get_best_mask(segmentation, 3)
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(segmentation_quality, 0.01)
new_params, _ = run_rank_pf(img, None, None, gt, cellstar.parameters)
cellstar = Segmentation(11, 20)
cellstar.parameters = new_params
cellstar.set_frame(img)
segmentation2, snakes2 = cellstar.run_segmentation()
self.assertLessEqual(3, segmentation2.max())
self.assertLessEqual(3, len(snakes2))
# find best 3 objects
best3 = get_best_mask(segmentation2, 3)
snake3 = snakes2[:3]
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(0.3, segmentation_quality) # there was no contour fitting
object_diffs = calculate_diffs_per_object(best3, gt)
self.assertLessEqual(0.3, min(object_diffs))
def test_contour_fitting(self):
process.SEARCH_LENGTH_NORMAL = 20
img = prepare_image((80, 80))
gt = np.zeros((80,80), dtype=int)
draw_weak_cell(img, (60, 20), 12)
draw_disc(gt, (60, 20), 14, 1)
draw_weak_cell(img, (40, 25), 8)
draw_disc(gt, (40, 25), 8, 2)
draw_weak_cell(img, (30, 40), 9)
draw_disc(gt, (30, 40), 9, 3)
img = finish_image(img)
cellstar = Segmentation(11, 20)
# break parameters
weights = cellstar.parameters["segmentation"]["stars"]["sizeWeight"]
encoded = params.pf_parameters_encode(cellstar.parameters)
one_params = [0.5 for k in encoded]
one_params[3] = 0.001
one_decoded = params.pf_parameters_decode(one_params, weights)
broken_params = PFSnake.merge_parameters(cellstar.parameters, one_decoded)
cellstar.parameters = broken_params
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
# fail miserably (not all snakes are found)
self.assertGreater(3, segmentation.max())
self.assertGreater(3, len(snakes))
new_params, _ = run_pf(img, None, None, gt, cellstar.parameters, 11, 20)
cellstar = Segmentation(11, 20)
cellstar.parameters = new_params
cellstar.set_frame(img)
segmentation2, snakes2 = cellstar.run_segmentation()
self.assertLessEqual(3, segmentation2.max())
self.assertLessEqual(3, len(snakes2))
# find best 3 objects
best3 = get_best_mask(segmentation2, 3)
segmentation_quality = calculate_diff_fraction(best3, gt)
self.assertLessEqual(0.7, segmentation_quality)
def test_trained_parameters(image,
star_params,
ranking_params,
precision,
avg_cell_diameter,
output_name=None):
seg = Segmentation(segmentation_precision=precision,
avg_cell_diameter=avg_cell_diameter)
for k, v in star_params.iteritems():
seg.parameters["segmentation"]["stars"][k] = v
for k, v in ranking_params.iteritems():
seg.parameters["segmentation"]["ranking"][k] = v
seg.set_frame(image)
seg.run_segmentation()
if output_name is None:
image_show(seg.images.segmentation, 1)
else:
image_save(seg.images.segmentation, output_name)
def test_large_image(self):
img = prepare_image((300, 300))
gt = np.zeros((300, 300), dtype=int)
draw_cell(img, (40, 40), 30)
draw_disc(gt, (40, 40), 30, 1)
draw_cell(img, (150, 90), 25)
draw_disc(gt, (150, 90), 25, 2)
draw_cell(img, (180, 70), 20)
draw_disc(gt, (180, 70), 20, 3)
cellstar = Segmentation(11, 60)
cellstar.set_frame(img)
segmentation, snakes = cellstar.run_segmentation()
self.assertEqual(3, segmentation.max())
self.assertEqual(3, len(snakes))
object_diffs = calculate_diffs_per_object(segmentation, gt)
self.assertLessEqual(0.8, min(object_diffs))