-
Notifications
You must be signed in to change notification settings - Fork 15
/
train.py
159 lines (140 loc) · 6.39 KB
/
train.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
"""
This code can be used to train NuClick on your own data.
You can train from scratch or fine-tuning one of the existing models.
Generally, training models follows the scheme of one of following applications:
Nucleus, Cell, Glands
However, you can control the training process by adjusting the parameters in the
config.py files.
For more information, read our arXiv paper:
NuClick: From Clicks in the Nuclei to Nuclear Boundaries
https://arxiv.org/abs/1909.03253
"""
from __future__ import print_function
import os
# os.environ["CUDA_VISIBLE_DEVICES"] = config.gpu
from skimage.io import imsave
import numpy as np
from keras.callbacks import ModelCheckpoint, CSVLogger
from keras import backend as K
from data_handler.npyDataOps import loadData
from data_handler.customImageGenerator import ImageDataGenerator
from config import config
import h5py
import warnings
from models.models import getModel
from utils.ModelCheckpointMGPU import ModelCheckpointMGPU
warnings.filterwarnings("ignore")
# Setting Parameters
SEED = 1
np.random.seed(SEED)
K.set_image_data_format('channels_last') # TF dimension ordering in this code
img_rows = config.img_rows # 480#640
img_cols = config.img_cols # 768#1024
img_chnls = config.img_chnls
input_shape = (img_rows, img_cols)
modelType = config.modelType
lossType = config.lossType
batchSize = config.batchSize
batchSizeVal = batchSize
gpus = [x.name for x in K.device_lib.list_local_devices() if x.name[:11] == '/device:GPU']
multiGPU = config.multiGPU
print('-' * 30)
print('Loading data...')
print('-' * 30)
imgs, masks, weightMaps, objectPoints, pointOthers, imgNames = loadData(config.train_data_path)
weightMaps = weightMaps[..., np.newaxis]
masks = masks[..., np.newaxis]
objectPoints = objectPoints[..., np.newaxis]
pointOthers = pointOthers[..., np.newaxis]
guidingSignals = np.concatenate((objectPoints,pointOthers,weightMaps),axis=3)
del weightMaps
del objectPoints
del pointOthers
print('Train data loading is done.')
if not config.valid_data_path==None:
imgs_test, masks_test, weightMaps_test, objectPoints_test, pointOthers_test, imgNames_test = loadData(config.valid_data_path)
weightMaps_test = weightMaps_test[..., np.newaxis]
masks_test = masks_test[..., np.newaxis]
objectPoints_test = objectPoints_test[..., np.newaxis]
pointOthers_test = pointOthers_test[..., np.newaxis]
guidingSignals_test = np.concatenate((objectPoints_test, pointOthers_test, weightMaps_test),axis=3)
del weightMaps_test
del objectPoints_test
del pointOthers_test
print('Test data loading is done.')
# Defining train/validation image generators and their argumants
train_gen_args = dict(
RandomizeGuidingSignalType = config.guidingSignalType,
width_shift_range=0.1,
height_shift_range=0.1,
horizontal_flip=True,
vertical_flip=True,
rotation_range=20,
zoom_range=(.5, 1.5) if config.application=='Gland' else (.7, 1.3),
shear_range=0.2 if config.application=='Gland' else 0.1,
fill_mode='constant', # Applicable to image onlyS
albumentation=True,
rescale=1. / 255)
image_datagen = ImageDataGenerator(**train_gen_args)
image_datagen_val = ImageDataGenerator(rescale=1./255)
if not config.valid_data_path==None:
num_train = imgs.shape[0] # 0
num_val = imgs_test.shape[0]
train_generator = image_datagen.flow( imgs, weightMap=guidingSignals, mask=masks,
shuffle=True, batch_size=batchSize, color_mode='rgb', seed=SEED)
val_generator = image_datagen_val.flow(imgs_test, weightMap=guidingSignals_test, mask=masks_test,
shuffle=False, batch_size=batchSizeVal, color_mode='rgb', seed=SEED)
else:
num_train = np.round((1-config.valPrec)*imgs.shape[0])
num_val = imgs.shape[0]-num_train
train_generator = image_datagen.flow(imgs[:num_train,], weightMap=guidingSignals[:num_train,], mask=masks[:num_train,],
shuffle=True, batch_size=batchSize, color_mode='rgb', seed=SEED)
val_generator = image_datagen_val.flow(imgs[num_train:,], weightMap=guidingSignals[num_train:,], mask=masks[num_train:,],
shuffle=False, batch_size=batchSizeVal, color_mode='rgb', seed=SEED)
imgNames_test = imgNames[num_train:,]
print('-' * 30)
print('Creating and compiling model...')
print('-' * 30)
# defining names and adresses
modelBaseName = 'NuClick_%s_%s_%s' % (config.application, modelType, lossType)
if not os.path.exists(config.weights_path):
os.mkdir(config.weights_path)
modelSaveName = "%s/weights-%s.h5" % (config.weights_path, modelBaseName)
modelLogName = "%s/Log-%s.log" % (config.weights_path, modelBaseName)
csv_logger = CSVLogger(modelLogName, append=True, separator='\t')
# creating model instance
model = getModel(modelType, lossType, input_shape)
if config.resumeTraining:
model.load_weights(modelSaveName)
if config.multiGPU:
model_checkpoint = ModelCheckpointMGPU(model, filepath=modelSaveName, monitor='val_loss', mode='min', save_best_only=True)
else:
model_checkpoint = ModelCheckpoint(filepath=modelSaveName, monitor='val_loss', save_best_only=True)
print('-' * 30)
print('Fitting model...')
print('-' * 30)
history = model.fit_generator(train_generator, steps_per_epoch=num_train//batchSize, nb_epoch=5,
validation_data=val_generator, validation_steps=num_val//batchSizeVal,
callbacks=[model_checkpoint, csv_logger], max_queue_size=64, workers=8)
history = model.fit_generator(train_generator, steps_per_epoch=num_train//batchSize, nb_epoch=150,
validation_data=val_generator, validation_steps=num_val//batchSizeVal,
callbacks=[model_checkpoint, csv_logger], max_queue_size=64, workers=8)
print('*' * 90)
print('Training is Done')
print('*' * 90)
if config.outputValPreds:
print('-' * 30)
print('Predicting on validation...')
print('-' * 30)
model.load_weights(modelSaveName)
batchSizeVal = 1
val_generator.shuffle=False
val_generator.batch_size=1
val_predicts = model.predict_generator(val_generator, steps=num_val//batchSizeVal)
pred_dir = "%s/valPred_%s" % (config.preds_path, modelBaseName)
if not os.path.exists(pred_dir):
os.makedirs(pred_dir, exist_ok=True)
imgs_mask_test = np.matrix.squeeze(val_predicts, axis=3)
for image_id in range(0, len(imgs_mask_test)):
mask = np.uint8(imgs_mask_test[image_id, :, :] * 255)
imsave(os.path.join(pred_dir, imgNames_test[image_id] + '_pred.png'), mask)