def test_tile_overlap(n_depth, kern_size):
K.clear_session()
img_size = 1280
rf_x, rf_y = receptive_field_unet(n_depth, kern_size, 2, img_size)
assert rf_x == rf_y
rf = rf_x
assert np.abs(rf[0] - rf[1]) < 10
assert sum(rf) + 1 < img_size
assert max(rf) == tile_overlap(n_depth, kern_size)
def test_tile_overlap():
n_depths = (1, 2, 3, 4, 5)
n_kernel = (3, 5, 7)
img_size = 1280
for k in n_kernel:
for n in n_depths:
K.clear_session()
rf_x, rf_y = receptive_field_unet(n, k, 2, img_size)
assert rf_x == rf_y
rf = rf_x
assert np.abs(rf[0] - rf[1]) < 10
assert sum(rf) + 1 < img_size
assert max(rf) == tile_overlap(n, k)
def get_yml_dict(self,
name,
description,
authors,
test_img,
axes,
patch_shape=None):
if (patch_shape != None):
self.config.patch_shape = patch_shape
''' Repeated values to avoid reference tags of the form &id002 in yml output when the same variable is used more than
once in the dictionary'''
mean_val = []
mean_val1 = []
for ele in self.config.means:
mean_val.append(float(ele))
mean_val1.append(float(ele))
std_val = []
std_val1 = []
for ele in self.config.stds:
std_val.append(float(ele))
std_val1.append(float(ele))
in_data_range_val = ['-inf', 'inf']
out_data_range_val = ['-inf', 'inf']
axes_val = 'b' + self.config.axes
axes_val = axes_val.lower()
val = 2**self.config.unet_n_depth
val1 = predict.tile_overlap(self.config.unet_n_depth,
self.config.unet_kern_size)
min_val = [1, val, val, self.config.n_channel_in]
step_val = [1, val, val, 0]
halo_val = [0, val1, val1, 0]
scale_val = [1, 1, 1, 1]
offset_val = [0, 0, 0, 3]
yaml = YAML(typ='rt')
with open(self.logdir / 'config.json', 'r') as f:
tr_kwargs_val = yaml.load(f)
if (self.config.n_dim == 3):
min_val = [1, val, val, val, self.config.n_channel_in]
step_val = [1, val, val, val, 0]
halo_val = [0, val1, val1, val1, 0]
scale_val = [1, 1, 1, 1, 1]
offset_val = [0, 0, 0, 0, 0]
yml_dict = {
'name':
name,
'description':
description,
'cite': [{
'text':
'Tim-Oliver Buchholz and Mangal Prakash and Alexander Krull and Florian Jug DenoiSeg: Joint Denoising and Segmentation\nArXiv (2020)',
'doi': 'arXiv:2005.02987'
}],
'authors':
authors,
'language':
'python',
'framework':
'tensorflow',
'format_version':
'0.2.0-csbdeep',
'source':
'denoiseg',
'test_input':
'testinput.tif',
'test_output':
'testoutput.tif',
'inputs': [{
'name': 'input',
'axes': axes_val,
'data_type': 'float32',
'data_range': in_data_range_val,
'halo': halo_val,
'shape': {
'min': min_val,
'step': step_val
}
}],
'outputs': [{
'name': self.keras_model.layers[-1].output.name,
'axes': axes_val,
'data_type': 'float32',
'data_range': out_data_range_val,
'shape': {
'reference_input': 'input',
'scale': scale_val,
'offset': offset_val
}
}],
'training': {
'source': 'n2v.train()',
'kwargs': tr_kwargs_val
},
'prediction': {
'weights': {
'source': './variables/variables'
},
'preprocess': [{
'kwargs': {
'mean': mean_val,
'stdDev': std_val
}
}],
'postprocess': [{
'kwargs': {
'mean': mean_val1,
'stdDev': std_val1
}
}]
}
}
return yml_dict
def get_yml_dict(self,
name,
description,
authors,
test_img,
axes,
patch_shape=None):
if (patch_shape != None):
self.config.patch_shape = patch_shape
''' Repeated values to avoid reference tags of the form &id002 in yml output when the same variable is used more than
once in the dictionary'''
mean_val = []
mean_val1 = []
for ele in self.config.means:
mean_val.append(float(ele))
mean_val1.append(float(ele))
std_val = []
std_val1 = []
for ele in self.config.stds:
std_val.append(float(ele))
std_val1.append(float(ele))
in_data_range_val = ['-inf', 'inf']
out_data_range_val = ['-inf', 'inf']
axes_val = 'b' + self.config.axes
axes_val = axes_val.lower()
val = 2**self.config.unet_n_depth
val1 = predict.tile_overlap(self.config.unet_n_depth,
self.config.unet_kern_size)
min_val = [1, val, val, self.config.n_channel_in]
step_val = [1, val, val, 0]
halo_val = [0, val1, val1, 0]
scale_val = [1, 1, 1, 1]
offset_val = [0, 0, 0, 0]
yaml = YAML(typ='rt')
with open(self.logdir / 'config.json', 'r') as f:
tr_kwargs_val = yaml.load(f)
if (self.config.n_dim == 3):
min_val = [1, val, val, val, self.config.n_channel_in]
step_val = [1, val, val, val, 0]
halo_val = [0, val1, val1, val1, 0]
scale_val = [1, 1, 1, 1, 1]
offset_val = [0, 0, 0, 0, 0]
yml_dict = {
'name':
name,
'description':
description,
'cite': [{
'text':
'Krull, A. and Buchholz, T. and Jug, F. Noise2void - learning denoising from single noisy images.\nProceedings of the IEEE Conference on Computer Vision and Pattern Recognition (2019)',
'doi': '10.1109/CVPR.2019.00223'
}],
'authors':
authors,
'language':
'python',
'framework':
'tensorflow',
'format_version':
'0.2.0-csbdeep',
'source':
'n2v',
'test_input':
'testinput.tif',
'test_output':
'testoutput.tif',
'inputs': [{
'name': 'input',
'axes': axes_val,
'data_type': 'float32',
'data_range': in_data_range_val,
'halo': halo_val,
'shape': {
'min': min_val,
'step': step_val
}
}],
'outputs': [{
'name': self.keras_model.layers[-1].output.name,
'axes': axes_val,
'data_type': 'float32',
'data_range': out_data_range_val,
'shape': {
'reference_input': 'input',
'scale': scale_val,
'offset': offset_val
}
}],
'training': {
'source': 'n2v.train()',
'kwargs': tr_kwargs_val
},
'prediction': {
'weights': {
'source': './variables/variables'
},
'preprocess': [{
'kwargs': {
'mean': mean_val,
'stdDev': std_val
}
}],
'postprocess': [{
'kwargs': {
'mean': mean_val1,
'stdDev': std_val1
}
}]
}
}
return yml_dict