class Segmentation():
def __init__(self,
img_shape=None,
pixel_size=None,
r_min=None,
frame_data=None,
edge_dist=15,
channel_width=0,
**kwargs):
# rmin in µm
# channel_width in pixel ?
# pixel_size in m / not in mµ!!
self.unet = UNet().create_model((img_shape[0], img_shape[1], 1),
1,
d=8)
network_path = "/home/user/Desktop/2020_Deformation_Cytometer/models_local/background_weighting_with_custom_function/Unet_andy_Unet_class_weight_like_cw0_01_20201110-041226.h5"
self.unet.load_weights(network_path)
self.pixel_size = pixel_size
self.r_min = r_min
self.frame_data = frame_data if frame_data is not frame_data else {}
self.edge_dist = edge_dist
self.config = {}
self.config["channel_width_px"] = channel_width
self.config["pixel_size_m"] = pixel_size
def segmentation(self, img):
if len(img.shape) == 3:
img = img[:, :, 0]
img = (img - np.mean(img)) / np.std(img).astype(np.float32)
prob_map = self.unet.predict(img[None, :, :, None])[0, :, :, 0]
prediction_mask = prob_map > 0.5
cells = mask_to_cells(prediction_mask, img, self.config, self.r_min,
self.frame_data, self.edge_dist)
return prediction_mask, cells, prob_map
class Segmentation():
def __init__(self,
img_shape=None,
pixel_size=None,
r_min=None,
frame_data=None,
edge_dist=15,
channel_width=0,
**kwargs):
# rmin in µm
# channel_width in pixel ?
# pixel_size in m / not in mµ!!
self.unet = UNet().create_model((img_shape[0], img_shape[1], 1),
1,
d=8)
network_path = "/home/user/Downloads/andyUnet_andy_bens_network_long_n300_20201027-082115_checkpoint.h5"
self.unet.load_weights(network_path)
self.pixel_size = pixel_size
self.r_min = r_min
self.frame_data = frame_data if frame_data is not frame_data else {}
self.edge_dist = edge_dist
self.config = {}
self.config["channel_width_px"] = channel_width
self.config["pixel_size_m"] = pixel_size
def segmentation(self, img):
if len(img.shape) == 3:
img = img[:, :, 0]
img = (img - np.mean(img)) / np.std(img).astype(np.float32)
prob_map = self.unet.predict(img[None, :, :, None])[0, :, :, 0]
prediction_mask = prob_map > 0.5
cells = mask_to_cells(prediction_mask, img, self.config, self.r_min,
self.frame_data, self.edge_dist)
return prediction_mask, cells, prob_map
class Segmentation():
def __init__(self,
img_shape=None,
pixel_size=None,
r_min=None,
frame_data=None,
edge_dist=15,
channel_width=0,
**kwargs):
# rmin in µm
# channel_width in pixel ?
# pixel_size in m / not in mµ!!
self.unet = UNet().create_model((img_shape[0], img_shape[1], 1),
1,
d=8)
network_path = "/home/user/Software/Deformation_Cytometer/deformationcytometer/detection/includes/v0.3/Unet_transfer_immune_cells.h5"
self.unet.load_weights(network_path)
self.pixel_size = pixel_size
self.r_min = r_min
self.frame_data = frame_data if frame_data is not frame_data else {}
self.edge_dist = edge_dist
self.config = {}
self.config["channel_width_px"] = channel_width
self.config["pixel_size_m"] = pixel_size
def segmentation(self, img):
if len(img.shape) == 3:
img = img[:, :, 0]
img = (img - np.mean(img)) / np.std(img).astype(np.float32)
prob_map = self.unet.predict(img[None, :, :, None])[0, :, :, 0]
prediction_mask = prob_map > 0.5
cells = mask_to_cells(prediction_mask, img, self.config, self.r_min,
self.frame_data, self.edge_dist)
return prediction_mask, cells, prob_map
class Segmentation():
def __init__(self,
img_shape=None,
pixel_size=None,
r_min=None,
frame_data=None,
edge_dist=15,
channel_width=0,
**kwargs):
# rmin in µm
# channel_width in pixel ?
# pixel_size in m / not in mµ!!
self.unet = UNet().create_model((img_shape[0], img_shape[1], 1),
1,
d=8)
network_path = "/home/user/Desktop/2020_Deformation_Cytometer/local_training/Unet_test_reproducability_1_20201203-104240.h5"
self.unet.load_weights(network_path)
self.pixel_size = pixel_size
self.r_min = r_min
self.frame_data = frame_data if frame_data is not frame_data else {}
self.edge_dist = edge_dist
self.config = {}
self.config["channel_width_px"] = channel_width
self.config["pixel_size_m"] = pixel_size
def segmentation(self, img):
if len(img.shape) == 3:
img = img[:, :, 0]
img = (img - np.mean(img)) / np.std(img).astype(np.float32)
prob_map = self.unet.predict(img[None, :, :, None])[0, :, :, 0]
prediction_mask = prob_map > 0.5
cells, prediction_mask = mask_to_cells_edge(prediction_mask,
img,
self.config,
self.r_min,
self.frame_data,
self.edge_dist,
return_mask=True)
return prediction_mask, cells, prob_map
class Segmentation():
def __init__(self,
network_path=None,
img_shape=None,
pixel_size=None,
r_min=None,
frame_data=None,
edge_dist=15,
channel_width=0,
edge_only=False,
return_mask=True,
d=8,
**kwargs):
# rmin in µm
# channel_width in pixel ?
# pixel_size in m / not in mµ!!
self.unet = UNet().create_model((img_shape[0], img_shape[1], 1),
1,
d=d)
#network_path ="/home/user/Desktop/2020_Deformation_Cytometer/models_local/weights_andy_transfer_learning1/andyUnet_andy_transfer_long_n200__20201006-155443.h5"
self.unet.load_weights(network_path)
self.pixel_size = pixel_size
self.r_min = r_min
self.frame_data = frame_data if frame_data is not frame_data else {}
self.edge_dist = edge_dist
self.config = {}
self.config["channel_width_px"] = channel_width
self.config["pixel_size_m"] = pixel_size
self.edge_only = edge_only
self.return_mask = return_mask
def search_cells(self, prediction_mask, img):
if self.edge_only:
if self.return_mask:
cells, prediction_mask = mask_to_cells_edge(
prediction_mask,
img,
self.config,
self.r_min,
self.frame_data,
self.edge_dist,
return_mask=self.return_mask)
else:
cells = mask_to_cells_edge(prediction_mask,
img,
self.config,
self.r_min,
self.frame_data,
self.edge_dist,
return_mask=self.return_mask)
else:
cells = mask_to_cells(prediction_mask, img, self.config,
self.r_min, self.frame_data, self.edge_dist)
return prediction_mask, cells
def segmentation(self, img):
# image batch
if len(img.shape) == 4:
img = preprocess_batch(img)
prediction_mask = self.unet.predict(img) > 0.5
cells = []
for i in range(prediction_mask.shape[0]):
_, cells_ = self.search_cells(prediction_mask[i, :, :, 0],
img[i, :, :, 0])
cells.extend(cells_)
prediction_mask = None
# single image
elif len(img.shape) == 2:
img = (img - np.mean(img)) / np.std(img).astype(np.float32)
prediction_mask = self.unet.predict(img[None, :, :, None])[0, :, :,
0] > 0.5
prediction_mask, cells = self.search_cells(prediction_mask, img)
else:
raise Exception("incorrect image shape: img.shape == " +
str(img.shape))
return prediction_mask, cells
