def NuclearSegmentationModel(input_shape=(None, None, 1),
backbone='resnet50',
use_pretrained_weights=True):
"""
A RetinaMask model with a ResNet50 backbone
for nuclear segmentation trained on DAPI data.
Args:
input_shape (tuple): a 3-length tuple of the input data shape.
backbone (str): name of the backbone to use for the model.
use_pretrained_weights (bool): whether to load pre-trained weights.
Only supports the ResNet50 backbone.
"""
backbone_levels = ['C1', 'C2', 'C3', 'C4', 'C5']
pyramid_levels = ['P2', 'P3', 'P4']
anchor_size_dict = {'P2': 8, 'P3': 16, 'P4': 32}
# Set up the prediction model
anchor_params = generate_anchor_params(pyramid_levels, anchor_size_dict)
model = model_zoo.RetinaMask(
backbone=backbone,
use_imagenet=False,
panoptic=True,
num_semantic_heads=2,
num_semantic_classes=[4, 4],
input_shape=input_shape,
num_classes=1,
backbone_levels=backbone_levels,
pyramid_levels=pyramid_levels,
anchor_params=anchor_params,
norm_method='whole_image')
if use_pretrained_weights:
if backbone == 'resnet50':
# '/data/models/resnet50_panoptic_train-val_DVV_V4.h5'
weights_path = get_file(
'resnet50_panoptic_nuclear_segmentation.h5',
WEIGHTS_PATH,
cache_subdir='models',
md5_hash='6e925c49cb05a1e3b0e2210220922445')
model.load_weights(weights_path)
else:
raise ValueError('Backbone %s does not have a weights file.' %
backbone)
return model
def PhaseSegmentationModel(input_shape=(None, None, 1),
backbone='resnet50',
use_pretrained_weights=True):
"""
A RetinaMask model with a ResNet50 backbone
for cytoplasmic segmentation trained on phase data.
Args:
input_shape (tuple): a 3-length tuple of the input data shape.
backbone (str): name of the backbone to use for the model.
use_pretrained_weights (bool): whether to load pre-trained weights.
Only supports the ResNet50 backbone.
"""
backbone_levels = ['C1', 'C2', 'C3', 'C4', 'C5']
pyramid_levels = ['P3', 'P4', 'P5', 'P6']
anchor_size_dicts = {'P3': 32, 'P4': 64, 'P5': 128, 'P6': 256}
anchor_params = generate_anchor_params(pyramid_levels, anchor_size_dicts)
model = model_zoo.RetinaMask(
backbone=backbone,
use_imagenet=False,
panoptic=False,
num_semantic_heads=2,
num_semantic_classes=[4, 4],
input_shape=input_shape,
num_classes=1,
backbone_levels=backbone_levels,
pyramid_levels=pyramid_levels,
anchor_params=anchor_params,
norm_method='whole_image')
if use_pretrained_weights:
if backbone == 'resnet50':
weights_path = get_file(
'resnet50_retinanet_20190813_all_phase_512.h5',
WEIGHTS_PATH,
cache_subdir='models',
md5_hash='ce31d084fadf7b907a25ab1fcf25529a')
model.load_weights(weights_path)
else:
raise ValueError('Backbone %s does not have a weights file.' %
backbone)
return model
def PhaseSegmentationModel(input_shape=(None, None, 1),
backbone='resnet50',
use_pretrained_weights=True):
"""Initialize a model for cytoplasmic segmentation based on phase data.
"""
backbone_levels = ['C1', 'C2', 'C3', 'C4', 'C5']
pyramid_levels = ['P3', 'P4', 'P5', 'P6']
anchor_size_dicts = {'P3': 32, 'P4': 64, 'P5': 128, 'P6': 256}
anchor_params = generate_anchor_params(pyramid_levels, anchor_size_dicts)
model = model_zoo.RetinaMask(backbone=backbone,
use_imagenet=False,
panoptic=False,
num_semantic_heads=2,
num_semantic_classes=[4, 4],
input_shape=input_shape,
num_classes=1,
backbone_levels=backbone_levels,
pyramid_levels=pyramid_levels,
anchor_params=anchor_params,
norm_method='whole_image')
if use_pretrained_weights:
if backbone == 'resnet50':
weights_path = get_file(
'resnet50_retinanet_20190813_all_phase_512.h5',
WEIGHTS_PATH,
cache_subdir='models',
md5_hash='ce31d084fadf7b907a25ab1fcf25529a')
model.load_weights(weights_path)
else:
raise ValueError('Backbone %s does not have a weights file.' %
backbone)
return model
def NuclearSegmentationModel(input_shape=(None, None, 1),
backbone='resnet50',
use_pretrained_weights=True):
"""Initialize a model for nuclear segmentation based on dapi data."""
backbone_levels = ['C1', 'C2', 'C3', 'C4', 'C5']
pyramid_levels = ['P2', 'P3', 'P4']
anchor_size_dict = {'P2': 8, 'P3': 16, 'P4': 32}
# Set up the prediction model
anchor_params = generate_anchor_params(pyramid_levels, anchor_size_dict)
model = model_zoo.RetinaMask(backbone=backbone,
use_imagenet=False,
panoptic=True,
num_semantic_heads=2,
num_semantic_classes=[4, 4],
input_shape=input_shape,
num_classes=1,
backbone_levels=backbone_levels,
pyramid_levels=pyramid_levels,
anchor_params=anchor_params,
norm_method='whole_image')
if use_pretrained_weights:
if backbone == 'resnet50':
# '/data/models/resnet50_panoptic_train-val_DVV_V4.h5'
weights_path = get_file(
'resnet50_panoptic_nuclear_segmentation.h5',
WEIGHTS_PATH,
cache_subdir='models',
md5_hash='6e925c49cb05a1e3b0e2210220922445')
model.load_weights(weights_path)
else:
raise ValueError('Backbone %s does not have a weights file.' %
backbone)
return model