def test_nuclear_app(self):
with self.cached_session():
model = PanopticNet(
'resnet50',
input_shape=(128, 128, 1),
norm_method='whole_image',
num_semantic_heads=2,
num_semantic_classes=[1, 1],
location=True,
include_top=True,
lite=True,
use_imagenet=False,
interpolation='bilinear')
app = NuclearSegmentation(model)
# test output shape
shape = app.model.output_shape
self.assertIsInstance(shape, list)
self.assertEqual(len(shape), 2)
self.assertEqual(len(shape[0]), 4)
self.assertEqual(len(shape[1]), 4)
# test predict
x = np.random.rand(1, 500, 500, 1)
y = app.predict(x)
self.assertEqual(x.shape, y.shape)
def test_nuclear_app(self):
with self.cached_session():
app = NuclearSegmentation(use_pretrained_weights=False)
# test output shape
shape = app.model.output_shape
self.assertIsInstance(shape, list)
self.assertEqual(len(shape), 2)
self.assertEqual(len(shape[0]), 4)
self.assertEqual(len(shape[1]), 4)
# test predict
x = np.random.rand(1, 500, 500, 1)
y = app.predict(x)
self.assertEqual(x.shape, y.shape)
def deepcell_segment(image: Array, compartment: str = None) -> Array:
from deepcell.applications import (
MultiplexSegmentation,
NuclearSegmentation,
CytoplasmSegmentation,
)
if compartment is None:
# If image has more than one channel i.e. CYX
# assume segmentation of both compartments, otherwise nuclear
compartment = "both" if image.shape[-1] > 1 else "nuclear"
if compartment == "nuclear":
app = NuclearSegmentation()
pred = app.predict(image).squeeze()
if compartment == "cytoplasm":
app = CytoplasmSegmentation()
pred = app.predict(image).squeeze()
if compartment == "both":
app = MultiplexSegmentation()
pred = app.predict(image, compartment=compartment).squeeze()
if len(pred.shape) == 4:
pred = resize(pred, image.shape[1:])
return pred
def deepcell_segmenter(self, input_img, cell_dia=None):
from deepcell.applications import NuclearSegmentation
app = NuclearSegmentation()
im = np.expand_dims(input_img, axis=-1)
im = np.expand_dims(im, axis=0)
masks1 = app.predict(im, image_mpp=cell_dia)
masks = np.squeeze(masks1)
boundary = find_boundaries(masks,
connectivity=1,
mode='thick',
background=0)
boundary_img = (255 * boundary).astype('uint8')
resized_bound = boundary_img
filled1 = ndimage.binary_fill_holes(resized_bound)
mask = (255 * filled1).astype('uint8') - resized_bound
boundary = resized_bound.astype('uint8')
return boundary, mask
class TestNuclearSegmentation(test.TestCase):
def setUp(self):
self.app = NuclearSegmentation(use_pretrained_weights=False)
def test_nuclear_app(self):
# Check shape parameters
shape = self.app.model.output_shape
self.assertIsInstance(shape, list)
self.assertEqual(len(shape), 3)
self.assertEqual(len(shape[0]), 4)
self.assertEqual(len(shape[1]), 4)
self.assertEqual(len(shape[2]), 4)
def test_predict(self):
x = np.random.rand(1, 500, 500, 1)
y = self.app.predict(x)
self.assertEqual(x.shape[:-1], y.shape)
def setUp(self):
self.app = NuclearSegmentation(use_pretrained_weights=False)