def create_patches(self, normalization=None):
for ch in self.train_channels:
n_images = len(
list((pathlib.Path(self.out_dir) / "train_data" / "raw" /
"CH_{}".format(ch) / "GT").glob("*.tif")))
print("-- Creating {} patches for channel: {}".format(
n_images * self.n_patches_per_image, ch))
raw_data = RawData.from_folder(
basepath=pathlib.Path(self.out_dir) / "train_data" / "raw" /
"CH_{}".format(ch),
source_dirs=["low"],
target_dir="GT",
axes=self.axes,
)
if normalization is not None:
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=self.patch_size,
n_patches_per_image=self.n_patches_per_image,
save_file=self.get_training_patch_path() /
"CH_{}_training_patches.npz".format(ch),
verbose=False,
normalization=normalization,
)
else:
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=self.patch_size,
n_patches_per_image=self.n_patches_per_image,
save_file=self.get_training_patch_path() /
"CH_{}_training_patches.npz".format(ch),
verbose=False,
)
plt.figure(figsize=(16, 4))
rand_sel = numpy.random.randint(low=0, high=len(X), size=6)
plot_some(X[rand_sel, 0],
Y[rand_sel, 0],
title_list=[range(6)],
cmap="gray")
plt.show()
print("Done")
return
def _create(img_size, img_axes, patch_size, patch_axes):
X, Y, XYaxes = create_patches(
raw_data=get_data(n_images, img_axes, img_size),
patch_size=patch_size,
patch_axes=patch_axes,
n_patches_per_image=n_patches_per_image,
)
assert len(X) == n_images * n_patches_per_image
assert np.allclose(X, Y, atol=1e-6)
if patch_axes is not None:
assert XYaxes == 'SC' + patch_axes.replace('C', '')
def _create(img_size,img_axes,patch_size,patch_axes):
U,V = (rng.uniform(size=(n_images,)+img_size) for _ in range(2))
X,Y,XYaxes = create_patches (
raw_data = RawData.from_arrays(U,V,img_axes),
patch_size = patch_size,
patch_axes = patch_axes,
n_patches_per_image = n_patches_per_image,
save_file = save_file
)
(_X,_Y), val_data, _XYaxes = load_training_data(save_file,verbose=True)
assert val_data is None
assert _XYaxes[-1 if backend_channels_last else 1] == 'C'
_X,_Y = (move_image_axes(u,fr=_XYaxes,to=XYaxes) for u in (_X,_Y))
assert np.allclose(X,_X,atol=1e-6)
assert np.allclose(Y,_Y,atol=1e-6)
assert set(XYaxes) == set(_XYaxes)
assert load_training_data(save_file,validation_split=0.5)[2] is not None
assert all(len(x)==3 for x in load_training_data(save_file,n_images=3)[0])
def generate_2D_patch_training_data(BaseDirectory,
SaveNpzDirectory,
SaveName,
patch_size=(512, 512),
n_patches_per_image=64,
transforms=None):
raw_data = RawData.from_folder(
basepath=BaseDirectory,
source_dirs=['Original'],
target_dir='BinaryMask',
axes='YX',
)
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=patch_size,
n_patches_per_image=n_patches_per_image,
transforms=transforms,
save_file=SaveNpzDirectory + SaveName,
)
def split_chunks(
data: List[np.ndarray], n_chunks: int, psf_as: List[np.ndarray],
psf_bs: List[np.ndarray]
) -> Tuple[np.ndarray, np.ndarray, str, np.ndarray, np.ndarray]:
"""
Extract chunks from RawData and normalize them
:param data: The data to extract the chunks form
:param n_chunks: The number of chunks to extract
:param psf_as: The list of PSF corresponding to the first channel of data
:param psf_bs: The list of PSF corresponding to the second channel of data
:return: (X chunks, Y chunks, axes, psf_as, psf_bs)
"""
def _normalize(patches_x, patches_y, x, y, mask, channel):
return patches_x, patches_y
X, Y, axes = create_patches(data, (64, 64, 64, 2),
n_patches_per_image=n_chunks,
patch_filter=None,
shuffle=False)
return X, Y, axes, np.repeat(psf_as, n_chunks, axis=0), np.repeat(psf_bs,
n_chunks,
axis=0),
def data_generation(data_path, axes, patch_size, data_name):
"""Generates training data for training CARE network. `RawData` object defines how to get the pairs of low/high SNR stacks and the semantics of each axis. We have two folders "noisy" and "clean" where corresponding low and high-SNR stacks are TIFF images with identical filenames.
Parameters
----------
data_path : str
Path of the input data containing 'noisy' and 'clean' folder
axes : str
Semantic order each axes
patch_size : tuple
Size of the patches to crop the images
data_name : str
Name of the .npz file containing the pairs of images.
"""
raw_data = RawData.from_folder(
basepath=data_path,
source_dirs=['noisy'],
target_dir='clean',
axes=axes,
)
# Patch size that is a power of two along XYZT, or at least divisible by 8.
# By convention, the variable name `X` (or `x`) refers to an input variable for a machine learning model, whereas `Y` (or `y`) indicates an output variable.
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=patch_size,
n_patches_per_image=100,
save_file='data/' + data_name + '.npz',
)
assert X.shape == Y.shape
print("shape of X,Y =", X.shape)
print("axes of X,Y =", XY_axes)
for image in tqdm(imgList, 'Reading img'):
imgArray.append(imread(os.path.join(train_dir, image)))
labelArray = []
for label in tqdm(labelList, 'Reading label'):
labelArray.append(imread(os.path.join(label_dir, label)))
print(imgArray[0].shape)
print(labelArray[0].shape)
raw_data = RawData.from_arrays(imgArray, labelArray, axes='YX')
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=(128, 128, 1),
patch_axes='YXC',
n_patches_per_image=25,
save_file=
'/mnt/AE3205C73205958D/Data/3dliver_local/pc_adult/2d_slices/imagesXY/image_full/mydata_128x128patch.npz'
)
(X, Y), (X_val, Y_val), axes = load_training_data(
'/mnt/AE3205C73205958D/Data/3dliver_local/pc_adult/2d_slices/imagesXY/image_full/mydata_128x128patch.npz',
validation_split=0.1,
verbose=True)
n = 10
print(X[n].shape)
fig = plt.figure()
pl1_x = fig.add_subplot(2, 5, 1)
pl1_x.imshow(X[n][..., 0])
pl1_y = fig.add_subplot(2, 5, 6)
def train(Training_source=".",
Training_target=".",
model_name="No_name",
model_path=".",
Visual_validation_after_training=True,
number_of_epochs=100,
patch_size=64,
number_of_patches=10,
Use_Default_Advanced_Parameters=True,
number_of_steps=300,
batch_size=32,
percentage_validation=15):
'''
Main function of the script. Train the model an save in model_path
Parameters
----------
Training_source : (str) Path to the noisy images
Training_target : (str) Path to the GT images
model_name : (str) name of the model
model_path : (str) path of the model
Visual_validation_after_training : (bool) Predict a random image after training
Number_of_epochs : (int) epochs
path_size : (int) patch sizes
number_of_patches : (int) number of patches for each image
User_Default_Advances_Parameters : (bool) Use default parameters for the training
number_of_steps : (int) number of steps
batch_size : (int) batch size
percentage_validation : (int) percentage validation
Return
-------
void
'''
OutputFile = Training_target + "/*.tif"
InputFile = Training_source + "/*.tif"
base = "/content/"
training_data = base + "/my_training_data.npz"
if (Use_Default_Advanced_Parameters):
print("Default advanced parameters enabled")
batch_size = 64
percentage_validation = 10
percentage = percentage_validation / 100
#here we check that no model with the same name already exist, if so delete
if os.path.exists(model_path + '/' + model_name):
shutil.rmtree(model_path + '/' + model_name)
# The shape of the images.
x = imread(InputFile)
y = imread(OutputFile)
print('Loaded Input images (number, width, length) =', x.shape)
print('Loaded Output images (number, width, length) =', y.shape)
print("Parameters initiated.")
# RawData Object
# This object holds the image pairs (GT and low), ensuring that CARE compares corresponding images.
# This file is saved in .npz format and later called when loading the trainig data.
raw_data = data.RawData.from_folder(basepath=base,
source_dirs=[Training_source],
target_dir=Training_target,
axes='CYX',
pattern='*.tif*')
X, Y, XY_axes = data.create_patches(raw_data,
patch_filter=None,
patch_size=(patch_size, patch_size),
n_patches_per_image=number_of_patches)
print('Creating 2D training dataset')
training_path = model_path + "/rawdata"
rawdata1 = training_path + ".npz"
np.savez(training_path, X=X, Y=Y, axes=XY_axes)
# Load Training Data
(X, Y), (X_val,
Y_val), axes = load_training_data(rawdata1,
validation_split=percentage,
verbose=True)
c = axes_dict(axes)['C']
n_channel_in, n_channel_out = X.shape[c], Y.shape[c]
#Show_patches(X,Y)
#Here we automatically define number_of_step in function of training data and batch size
if (Use_Default_Advanced_Parameters):
number_of_steps = int(X.shape[0] / batch_size) + 1
print(number_of_steps)
#Here we create the configuration file
config = Config(axes,
n_channel_in,
n_channel_out,
probabilistic=False,
train_steps_per_epoch=number_of_steps,
train_epochs=number_of_epochs,
unet_kern_size=5,
unet_n_depth=3,
train_batch_size=batch_size,
train_learning_rate=0.0004)
print(config)
vars(config)
# Compile the CARE model for network training
model_training = CARE(config, model_name, basedir=model_path)
if (Visual_validation_after_training):
Cell_executed = 1
import time
start = time.time()
#@markdown ##Start Training
# Start Training
history = model_training.train(X, Y, validation_data=(X_val, Y_val))
print("Training, done.")
if (Visual_validation_after_training):
Predict_a_image(Training_source, Training_target, model_path,
model_training)
# Displaying the time elapsed for training
dt = time.time() - start
min, sec = divmod(dt, 60)
hour, min = divmod(min, 60)
print("Time elapsed:", hour, "hour(s)", min, "min(s)", round(sec),
"sec(s)")
Show_loss_function(history, model_path)
basepath=
'/run/user/1000/gvfs/smb-share:server=isiserver.curie.net,share=u934/equipe_bellaiche/l_sancere/Training_Data_Sets/Training_CARE_restoration/SpinwideFRAP4_Training_CARE_40x_bin2_reduced',
source_dirs=['Low'],
target_dir='GT',
axes='ZYX',
pattern='*.TIF')
# In[3]:
patch_size = (16, 64, 64)
n_patches_per_image = 64
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=
patch_size, #for bin1 it is 16 128 128 and for bin2 it is 16 64 64
n_patches_per_image=n_patches_per_image, #at least 64?
save_file=
'/run/media/sancere/DATA1/Lucas_NextonCreated_npz/Training_CARE_restoration_SpinwideFRAP4_Bin2_Reduced.npz',
)
# In[4]:
ConfigNPZ = open(
"/run/media/sancere/DATA1/Lucas_NextonCreated_npz/Parameters_Npz/ConfigNPZ_Training_CARE_restoration_SpinwideFRAP4_Bin2.txt",
"w+")
ConfigNPZ.write("patch_size = {} \n n_patches_per_image = {}".format(
patch_size, n_patches_per_image))
ConfigNPZ.close()
# In[5]:
def Train(self):
BinaryName = 'BinaryMask/'
RealName = 'RealMask/'
Raw = sorted(glob.glob(self.BaseDir + '/Raw/' + '*.tif'))
Path(self.BaseDir + '/' + BinaryName).mkdir(exist_ok=True)
Path(self.BaseDir + '/' + RealName).mkdir(exist_ok=True)
RealMask = sorted(glob.glob(self.BaseDir + '/' + RealName + '*.tif'))
ValRaw = sorted(glob.glob(self.BaseDir + '/ValRaw/' + '*.tif'))
ValRealMask = sorted(
glob.glob(self.BaseDir + '/ValRealMask/' + '*.tif'))
print('Instance segmentation masks:', len(RealMask))
if len(RealMask) == 0:
print('Making labels')
Mask = sorted(glob.glob(self.BaseDir + '/' + BinaryName + '*.tif'))
for fname in Mask:
image = imread(fname)
Name = os.path.basename(os.path.splitext(fname)[0])
Binaryimage = label(image)
imwrite((self.BaseDir + '/' + RealName + Name + '.tif'),
Binaryimage.astype('uint16'))
Mask = sorted(glob.glob(self.BaseDir + '/' + BinaryName + '*.tif'))
print('Semantic segmentation masks:', len(Mask))
if len(Mask) == 0:
print('Generating Binary images')
RealfilesMask = sorted(
glob.glob(self.BaseDir + '/' + RealName + '*tif'))
for fname in RealfilesMask:
image = imread(fname)
Name = os.path.basename(os.path.splitext(fname)[0])
Binaryimage = image > 0
imwrite((self.BaseDir + '/' + BinaryName + Name + '.tif'),
Binaryimage.astype('uint16'))
if self.GenerateNPZ:
raw_data = RawData.from_folder(
basepath=self.BaseDir,
source_dirs=['Raw/'],
target_dir='BinaryMask/',
axes='ZYX',
)
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=(self.PatchZ, self.PatchY, self.PatchX),
n_patches_per_image=self.n_patches_per_image,
save_file=self.BaseDir + self.NPZfilename + '.npz',
)
# Training UNET model
if self.TrainUNET:
print('Training UNET model')
load_path = self.BaseDir + self.NPZfilename + '.npz'
(X, Y), (X_val,
Y_val), axes = load_training_data(load_path,
validation_split=0.1,
verbose=True)
c = axes_dict(axes)['C']
n_channel_in, n_channel_out = X.shape[c], Y.shape[c]
config = Config(axes,
n_channel_in,
n_channel_out,
unet_n_depth=self.depth,
train_epochs=self.epochs,
train_batch_size=self.batch_size,
unet_n_first=self.startfilter,
train_loss='mse',
unet_kern_size=self.kern_size,
train_learning_rate=self.learning_rate,
train_reduce_lr={
'patience': 5,
'factor': 0.5
})
print(config)
vars(config)
model = CARE(config,
name='UNET' + self.model_name,
basedir=self.model_dir)
if self.copy_model_dir is not None:
if os.path.exists(self.copy_model_dir + 'UNET' +
self.copy_model_name + '/' +
'weights_now.h5') and os.path.exists(
self.model_dir + 'UNET' +
self.model_name + '/' +
'weights_now.h5') == False:
print('Loading copy model')
model.load_weights(self.copy_model_dir + 'UNET' +
self.copy_model_name + '/' +
'weights_now.h5')
if os.path.exists(self.model_dir + 'UNET' + self.model_name + '/' +
'weights_now.h5'):
print('Loading checkpoint model')
model.load_weights(self.model_dir + 'UNET' + self.model_name +
'/' + 'weights_now.h5')
if os.path.exists(self.model_dir + 'UNET' + self.model_name + '/' +
'weights_last.h5'):
print('Loading checkpoint model')
model.load_weights(self.model_dir + 'UNET' + self.model_name +
'/' + 'weights_last.h5')
if os.path.exists(self.model_dir + 'UNET' + self.model_name + '/' +
'weights_best.h5'):
print('Loading checkpoint model')
model.load_weights(self.model_dir + 'UNET' + self.model_name +
'/' + 'weights_best.h5')
history = model.train(X, Y, validation_data=(X_val, Y_val))
print(sorted(list(history.history.keys())))
plt.figure(figsize=(16, 5))
plot_history(history, ['loss', 'val_loss'],
['mse', 'val_mse', 'mae', 'val_mae'])
if self.TrainSTAR:
print('Training StarDistModel model with', self.backbone,
'backbone')
self.axis_norm = (0, 1, 2)
if self.CroppedLoad == False:
assert len(Raw) > 1, "not enough training data"
print(len(Raw))
rng = np.random.RandomState(42)
ind = rng.permutation(len(Raw))
X_train = list(map(ReadFloat, Raw))
Y_train = list(map(ReadInt, RealMask))
self.Y = [
label(DownsampleData(y, self.DownsampleFactor))
for y in tqdm(Y_train)
]
self.X = [
normalize(DownsampleData(x, self.DownsampleFactor),
1,
99.8,
axis=self.axis_norm) for x in tqdm(X_train)
]
n_val = max(1, int(round(0.15 * len(ind))))
ind_train, ind_val = ind[:-n_val], ind[-n_val:]
self.X_val, self.Y_val = [self.X[i] for i in ind_val
], [self.Y[i] for i in ind_val]
self.X_trn, self.Y_trn = [self.X[i] for i in ind_train
], [self.Y[i] for i in ind_train]
print('number of images: %3d' % len(self.X))
print('- training: %3d' % len(self.X_trn))
print('- validation: %3d' % len(self.X_val))
if self.CroppedLoad:
self.X_trn = self.DataSequencer(Raw,
self.axis_norm,
Normalize=True,
labelMe=False)
self.Y_trn = self.DataSequencer(RealMask,
self.axis_norm,
Normalize=False,
labelMe=True)
self.X_val = self.DataSequencer(ValRaw,
self.axis_norm,
Normalize=True,
labelMe=False)
self.Y_val = self.DataSequencer(ValRealMask,
self.axis_norm,
Normalize=False,
labelMe=True)
self.train_sample_cache = False
print(Config3D.__doc__)
anisotropy = (1, 1, 1)
rays = Rays_GoldenSpiral(self.n_rays, anisotropy=anisotropy)
if self.backbone == 'resnet':
conf = Config3D(
rays=rays,
anisotropy=anisotropy,
backbone=self.backbone,
train_epochs=self.epochs,
train_learning_rate=self.learning_rate,
resnet_n_blocks=self.depth,
train_checkpoint=self.model_dir + self.model_name + '.h5',
resnet_kernel_size=(self.kern_size, self.kern_size,
self.kern_size),
train_patch_size=(self.PatchZ, self.PatchX, self.PatchY),
train_batch_size=self.batch_size,
resnet_n_filter_base=self.startfilter,
train_dist_loss='mse',
grid=(1, 1, 1),
use_gpu=self.use_gpu,
n_channel_in=1)
if self.backbone == 'unet':
conf = Config3D(
rays=rays,
anisotropy=anisotropy,
backbone=self.backbone,
train_epochs=self.epochs,
train_learning_rate=self.learning_rate,
unet_n_depth=self.depth,
train_checkpoint=self.model_dir + self.model_name + '.h5',
unet_kernel_size=(self.kern_size, self.kern_size,
self.kern_size),
train_patch_size=(self.PatchZ, self.PatchX, self.PatchY),
train_batch_size=self.batch_size,
unet_n_filter_base=self.startfilter,
train_dist_loss='mse',
grid=(1, 1, 1),
use_gpu=self.use_gpu,
n_channel_in=1,
train_sample_cache=False)
print(conf)
vars(conf)
Starmodel = StarDist3D(conf,
name=self.model_name,
basedir=self.model_dir)
print(
Starmodel._axes_tile_overlap('ZYX'),
os.path.exists(self.model_dir + self.model_name + '/' +
'weights_now.h5'))
if self.copy_model_dir is not None:
if os.path.exists(self.copy_model_dir + self.copy_model_name +
'/' + 'weights_now.h5') and os.path.exists(
self.model_dir + self.model_name + '/' +
'weights_now.h5') == False:
print('Loading copy model')
Starmodel.load_weights(self.copy_model_dir +
self.copy_model_name + '/' +
'weights_now.h5')
if os.path.exists(self.copy_model_dir + self.copy_model_name +
'/' + 'weights_last.h5') and os.path.exists(
self.model_dir + self.model_name + '/' +
'weights_last.h5') == False:
print('Loading copy model')
Starmodel.load_weights(self.copy_model_dir +
self.copy_model_name + '/' +
'weights_last.h5')
if os.path.exists(self.copy_model_dir + self.copy_model_name +
'/' + 'weights_best.h5') and os.path.exists(
self.model_dir + self.model_name + '/' +
'weights_best.h5') == False:
print('Loading copy model')
Starmodel.load_weights(self.copy_model_dir +
self.copy_model_name + '/' +
'weights_best.h5')
if os.path.exists(self.model_dir + self.model_name + '/' +
'weights_now.h5'):
print('Loading checkpoint model')
Starmodel.load_weights(self.model_dir + self.model_name + '/' +
'weights_now.h5')
if os.path.exists(self.model_dir + self.model_name + '/' +
'weights_last.h5'):
print('Loading checkpoint model')
Starmodel.load_weights(self.model_dir + self.model_name + '/' +
'weights_last.h5')
if os.path.exists(self.model_dir + self.model_name + '/' +
'weights_best.h5'):
print('Loading checkpoint model')
Starmodel.load_weights(self.model_dir + self.model_name + '/' +
'weights_best.h5')
historyStar = Starmodel.train(self.X_trn,
self.Y_trn,
validation_data=(self.X_val,
self.Y_val),
epochs=self.epochs)
print(sorted(list(historyStar.history.keys())))
plt.figure(figsize=(16, 5))
plot_history(historyStar, ['loss', 'val_loss'], [
'dist_relevant_mae', 'val_dist_relevant_mae',
'dist_relevant_mse', 'val_dist_relevant_mse'
])
# axes = 'CYX',
# )
raw_data = RawData.from_arrays(X=[x], Y=[x], axes='CYX')
anisotropic_transform = anisotropic_distortions(
subsample=1,
psf=np.ones((3)) / 9, # use the actual PSF here
psf_axes='Y',
# poisson_noise = True,
# gauss_sigma = 0.1
)
X, Y, XY_axes = create_patches(
raw_data=raw_data,
patch_size=(x.shape[0], x.shape[1], x.shape[2]),
n_patches_per_image=1,
transforms=[anisotropic_transform],
)
# z = axes_dict(XY_axes)['Z']
# X = np.take(X,0,axis=z)
# Y = np.take(Y,0,axis=z)
# XY_axes = XY_axes.replace('Z','')
# assert X.shape == Y.shape
# print("shape of X,Y =", X.shape)
# print("axes of X,Y =", XY_axes)
# plt.figure(figsize=(389/100, 389/100))
for i in range(len(X)):
fig = plt.figure(frameon=False)
fig.set_size_inches(X[i].shape[1] / 200, X[i].shape[0] / 200)
ax = plt.Axes(fig, [0., 0., 1., 1.])
ax.set_axis_off()
from argparse import ArgumentParser
from csbdeep.data import RawData, create_patches
args = ArgumentParser()
args.add_argument("input_basepath")
args.add_argument("input_x")
args.add_argument("input_y")
args.add_argument("output_file")
args.add_argument("--patch_size_xy", default=32)
args.add_argument("--patch_size_z", default=16)
args.add_argument("--n_patches_per_image", default=750)
args.add_argument("--axes", default="ZYX")
args = args.parse_args()
data = RawData.from_folder(basepath=args.input_basepath,
source_dirs=[args.input_x],
target_dir=args.input_y,
axes=args.axes)
ps_xy = int(args.patch_size_xy)
ps_z = int(args.patch_size_z)
X, Y, axes = create_patches(data,
patch_size=(ps_z, ps_xy, ps_xy, 1),
n_patches_per_image=int(args.n_patches_per_image),
save_file=args.output_file,
patch_axes=args.axes + "C")
def original():
x = np.load('data/img006_noconv.npy')
x = np.moveaxis(x, 1, 2)
## initial axes are TZCYX, but we
axes = 'CZYX'
mypath = Path('isonet_psf_1')
mypath.mkdir(exist_ok=True)
subsample = 5.0 #10.2
print('image size =', x.shape)
print('image axes =', axes)
print('subsample factor =', subsample)
plt.switch_backend('agg')
if False:
plt.figure(figsize=(15, 15))
plot_some(np.moveaxis(x[0], 1, -1)[[5, -5]],
title_list=[['xy slice', 'xy slice']],
pmin=2,
pmax=99.8)
plt.savefig(mypath / 'datagen_1.png')
plt.figure(figsize=(15, 15))
plot_some(np.moveaxis(np.moveaxis(x[0], 1, -1)[:, [50, -50]], 1, 0),
title_list=[['xz slice', 'xz slice']],
pmin=2,
pmax=99.8,
aspect=subsample)
plt.savefig(mypath / 'datagen_2.png')
def gimmeit_gen():
## iterate over time dimension
for i in range(x.shape[0]):
yield x[i], x[i], axes, None
raw_data = RawData(gimmeit_gen, x.shape[0], "this is great!")
## initial idea
if False:
def buildkernel():
kern = np.exp(-(np.arange(10)**2 / 2))
kern /= kern.sum()
kern = kern.reshape([1, 1, -1, 1])
kern = np.stack([kern, kern], axis=1)
return kern
psf_kern = buildkernel()
## use Martins theoretical psf
if False:
psf_aniso = imread('data/psf_aniso_NA_0.8.tif')
psf_channels = np.stack([
psf_aniso,
] * 2, axis=1)
def buildkernel():
kernel = np.zeros(20)
kernel[7:13] = 1 / 6
## reshape into CZYX. long axis is X.
kernel = kernel.reshape([1, 1, -1])
## repeate same kernel for both channels
kernel = np.stack([kernel, kernel], axis=0)
return kernel
psf = buildkernel()
print(psf.shape)
## use theoretical psf
if False:
psf_channels = np.load('data/measured_psfs.npy')
psf = rotate(psf_channels, 90, axes=(1, 3))
iso_transform = data.anisotropic_distortions(
subsample=subsample,
psf=psf,
)
X, Y, XY_axes = data.create_patches(
raw_data=raw_data,
patch_size=(2, 1, 128, 128),
n_patches_per_image=256,
transforms=[iso_transform],
)
assert X.shape == Y.shape
print("shape of X,Y =", X.shape)
print("axes of X,Y =", XY_axes)
# remove dummy z dim to obtain multi-channel 2D patches
X = X[:, :, 0, ...]
Y = Y[:, :, 0, ...]
XY_axes = XY_axes.replace('Z', '')
assert X.shape == Y.shape
print("shape of X,Y =", X.shape)
print("axes of X,Y =", XY_axes)
np.savez(mypath / 'my_training_data.npz', X=X, Y=Y, axes=XY_axes)
for i in range(2):
plt.figure(figsize=(16, 4))
sl = slice(8 * i, 8 * (i + 1))
plot_some(np.moveaxis(X[sl], 1, -1),
np.moveaxis(Y[sl], 1, -1),
title_list=[np.arange(sl.start, sl.stop)])
plt.savefig(mypath / 'datagen_panel_{}.png'.format(i))
y = imread(
os.path.join(args.base_dir, "high_snr", f"img_{img_number}.tif"))
print(f"image size: {x.shape}")
plt.figure(figsize=(16, 10))
plot_some(
np.stack([x, y]),
title_list=[['low snr', 'high snr']],
)
plt.show()
if not os.path.exists(os.path.join(args.base_dir, "test")):
os.mkdir(os.path.join(args.base_dir, "test"))
shutil.move(
os.path.join(args.base_dir, "low_snr", f"img_{img_number}.tif"),
os.path.join(args.base_dir, "test", "low_snr.tif"))
shutil.move(
os.path.join(args.base_dir, "high_snr", f"img_{img_number}.tif"),
os.path.join(args.base_dir, "test", "high_snr.tif"))
# Read the pairs, passing in the axis semantics
raw_data = RawData.from_folder(basepath=args.base_dir,
source_dirs=['low_snr'],
target_dir='high_snr',
axes='YX')
# From the stacks, generate 2D patches, and save to the output_filename
create_patches(raw_data=raw_data,
patch_size=(128, 128),
n_patches_per_image=512,
save_file=os.path.join(args.base_dir, args.output_filename))
def training(self):
#Loading Files and Plot examples
basepath = 'data/'
training_original_dir = 'training/original/'
training_ground_truth_dir = 'training/ground_truth/'
validation_original_dir = 'validation/original/'
validation_ground_truth_dir = 'validation/ground_truth/'
import glob
from skimage import io
from matplotlib import pyplot as plt
training_original_files = sorted(
glob.glob(basepath + training_original_dir + '*.tif'))
training_original_file = io.imread(training_original_files[0])
training_ground_truth_files = sorted(
glob.glob(basepath + training_ground_truth_dir + '*.tif'))
training_ground_truth_file = io.imread(training_ground_truth_files[0])
print("Training dataset's number of files and dimensions: ",
len(training_original_files), training_original_file.shape,
len(training_ground_truth_files),
training_ground_truth_file.shape)
training_size = len(training_original_file)
validation_original_files = sorted(
glob.glob(basepath + validation_original_dir + '*.tif'))
validation_original_file = io.imread(validation_original_files[0])
validation_ground_truth_files = sorted(
glob.glob(basepath + validation_ground_truth_dir + '*.tif'))
validation_ground_truth_file = io.imread(
validation_ground_truth_files[0])
print("Validation dataset's number of files and dimensions: ",
len(validation_original_files), validation_original_file.shape,
len(validation_ground_truth_files),
validation_ground_truth_file.shape)
validation_size = len(validation_original_file)
if training_size == validation_size:
size = training_size
else:
print(
'Training and validation images should be of the same dimensions!'
)
plt.figure(figsize=(16, 4))
plt.subplot(141)
plt.imshow(training_original_file)
plt.subplot(142)
plt.imshow(training_ground_truth_file)
plt.subplot(143)
plt.imshow(validation_original_file)
plt.subplot(144)
plt.imshow(validation_ground_truth_file)
#preparing inputs for NN from pairs of 32bit TIFF image files with intensities in range 0..1. . Run it only once for each new dataset
from csbdeep.data import RawData, create_patches, no_background_patches
training_data = RawData.from_folder(
basepath=basepath,
source_dirs=[training_original_dir],
target_dir=training_ground_truth_dir,
axes='YX',
)
validation_data = RawData.from_folder(
basepath=basepath,
source_dirs=[validation_original_dir],
target_dir=validation_ground_truth_dir,
axes='YX',
)
# pathces will be created further in "data augmentation" step,
# that's why patch size here is the dimensions of images and number of pathes per image is 1
size1 = 64
X, Y, XY_axes = create_patches(
raw_data=training_data,
patch_size=(size1, size1),
patch_filter=no_background_patches(0),
n_patches_per_image=1,
save_file=basepath + 'training.npz',
)
X_val, Y_val, XY_axes = create_patches(
raw_data=validation_data,
patch_size=(size1, size1),
patch_filter=no_background_patches(0),
n_patches_per_image=1,
save_file=basepath + 'validation.npz',
)
#Loading training and validation data into memory
from csbdeep.io import load_training_data
(X, Y), _, axes = load_training_data(basepath + 'training.npz',
verbose=False)
(X_val,
Y_val), _, axes = load_training_data(basepath + 'validation.npz',
verbose=False)
X.shape, Y.shape, X_val.shape, Y_val.shape
from csbdeep.utils import axes_dict
c = axes_dict(axes)['C']
n_channel_in, n_channel_out = X.shape[c], Y.shape[c]
batch = len(
X
) # You should define number of batches according to the available memory
#batch=1
seed = 1
from keras.preprocessing.image import ImageDataGenerator
data_gen_args = dict(samplewise_center=False,
samplewise_std_normalization=False,
zca_whitening=False,
fill_mode='reflect',
rotation_range=30,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,
vertical_flip=True
#width_shift_range=0.2,
#height_shift_range=0.2,
)
# training
image_datagen = ImageDataGenerator(**data_gen_args)
mask_datagen = ImageDataGenerator(**data_gen_args)
image_datagen.fit(X, augment=True, seed=seed)
mask_datagen.fit(Y, augment=True, seed=seed)
image_generator = image_datagen.flow(X, batch_size=batch, seed=seed)
mask_generator = mask_datagen.flow(Y, batch_size=batch, seed=seed)
generator = zip(image_generator, mask_generator)
# validation
image_datagen_val = ImageDataGenerator(**data_gen_args)
mask_datagen_val = ImageDataGenerator(**data_gen_args)
image_datagen_val.fit(X_val, augment=True, seed=seed)
mask_datagen_val.fit(Y_val, augment=True, seed=seed)
image_generator_val = image_datagen_val.flow(X_val,
batch_size=batch,
seed=seed)
mask_generator_val = mask_datagen_val.flow(Y_val,
batch_size=batch,
seed=seed)
generator_val = zip(image_generator_val, mask_generator_val)
# plot examples
x, y = generator.__next__()
x_val, y_val = generator_val.__next__()
plt.figure(figsize=(16, 4))
plt.subplot(141)
plt.imshow(x[0, :, :, 0])
plt.subplot(142)
plt.imshow(y[0, :, :, 0])
plt.subplot(143)
plt.imshow(x_val[0, :, :, 0])
plt.subplot(144)
plt.imshow(y_val[0, :, :, 0])
import os
blocks = 2
channels = 16
learning_rate = 0.0004
learning_rate_decay_factor = 0.95
epoch_size_multiplicator = 20 # basically, how often do we decrease learning rate
epochs = 10
#comment='_side' # adds to model_name
import datetime
#model path
model_path = f'models/CSBDeep/model.h5'
if os.path.isfile(model_path):
print('Your model will be overwritten in the next cell')
kernel_size = 3
from csbdeep.models import Config, CARE
from keras import backend as K
best_mae = 1
steps_per_epoch = len(X) * epoch_size_multiplicator
validation_steps = len(X_val) * epoch_size_multiplicator
if 'model' in globals():
del model
if os.path.isfile(model_path):
os.remove(model_path)
for i in range(epochs):
print('Epoch:', i + 1)
learning_rate = learning_rate * learning_rate_decay_factor
config = Config(axes,
n_channel_in,
n_channel_out,
unet_kern_size=kernel_size,
train_learning_rate=learning_rate,
unet_n_depth=blocks,
unet_n_first=channels)
model = CARE(config, '.', basedir='models')
#os.remove('models/config.json')
if i > 0:
model.keras_model.load_weights(model_path)
model.prepare_for_training()
history = model.keras_model.fit_generator(
generator,
validation_data=generator_val,
validation_steps=validation_steps,
epochs=1,
verbose=0,
shuffle=True,
steps_per_epoch=steps_per_epoch)
if history.history['val_mae'][0] < best_mae:
best_mae = history.history['val_mae'][0]
if not os.path.exists('models/'):
os.makedirs('models/')
model.keras_model.save(model_path)
print(f'Validation MAE:{best_mae:.3E}')
del model
K.clear_session()
#model path
model_path = 'models/CSBDeep/model.h5'
config = Config(axes,
n_channel_in,
n_channel_out,
unet_kern_size=kernel_size,
train_learning_rate=learning_rate,
unet_n_depth=blocks,
unet_n_first=channels)
model = CARE(config, '.', basedir='models')
model.keras_model.load_weights(model_path)
model.export_TF()
