-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
226 lines (208 loc) · 8.3 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
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
# PARAMETERS
learning_rate = 0.001
batch_size = 10
epochs = 2
train_iterations = 100
use_gpu = True
import logging
encoding = "utf-8-sig" # or utf-8
logging.basicConfig(level=logging.INFO)
logger = logging.getLogger(__name__)
handler = logging.FileHandler('logger.log', encoding=encoding)
handler.setLevel(logging.INFO)
formatter = logging.Formatter('%(asctime)s - %(name)s - %(levelname)s - %(message)s')
handler.setFormatter(formatter)
logger.addHandler(handler)
logger.info('-' * 20 + 'START OF RUN' + '-' * 20)
# -*- coding: utf-8 -*-
logger.info('Loading modules')
import tensorflow as tf
config = tf.ConfigProto()
config.gpu_options.allow_growth = True # Stops tf allocating all memory
session = tf.Session(config=config)
# Sets keras backend to tensorflow
import os
os.environ['KERAS_BACKEND'] = 'tensorflow'
if not use_gpu:
os.environ["CUDA_VISIBLE_DEVICES"] = "-1"
import cv2
import numpy as np
import glob
import keras
import json
import os
import random
from keras.models import Sequential, load_model
from keras.layers import Dense, Dropout, Flatten, Embedding, Activation, BatchNormalization
from keras.layers import Convolution1D, Convolution2D, MaxPooling2D, AveragePooling2D, LSTM
from keras.layers import GlobalAveragePooling2D, UpSampling2D, TimeDistributed, Reshape
from keras.callbacks import ModelCheckpoint, TensorBoard
from tensorflow.python.client import device_lib
from keras.utils import np_utils
from keras.optimizers import SGD, Adam, RMSprop
import pdb
import time
import matplotlib
import re
import multiprocessing
import itertools
# Needed to change plot position while calculating. NEEDS TO ADDED BEFORE pyplot import
matplotlib.use('TkAgg')
import matplotlib.pyplot as plt
def log_func(txt):
logger.info(txt)
def main():
model = create_model()
model = compile_model(model)
all_groups = group_images()
random.shuffle(all_groups) # Shuffle data set
pool = multiprocessing.Pool(multiprocessing.cpu_count())
# NOTE: bad way of doing it. Spikes at 32 Gb or RAM. Perhaps a generator would be better?
all_training_data = pool.map(preprocess_data,
list_chunk(all_groups,
multiprocessing.cpu_count()))
pool.close()
all_training_data = list(itertools.chain(*all_training_data))
fit_model(model, all_training_data)
def list_chunk(l, n_chunks):
avg = len(l) / float(n_chunks)
out = []
last = 0.0
while last < len(l):
out.append(l[int(last):int(last + avg)])
last += avg
return out
def preprocess_data(all_groups):
all_training_data = []
for idx, group in enumerate(all_groups):
training_data = []
img, mask, percent = group
img_array = img_to_array(img)
if img_array.shape != (1024, 1024, 3):
# Note: Perhaps we could make it so that the CNN takes any size, but the output size would be all over the place
logger.warning(f'Image {img} has shape {img_array.shape}, not shape (1024, 1024, 3). Skipping')
continue
mask_array = normalize_mask(mask)
training_data.extend([img_array, mask_array, percent])
all_training_data.append(training_data)
return all_training_data
def normalize_mask(mask_file):
mask = cv2.imread(mask_file)
mask = cv2.cvtColor(mask, cv2.COLOR_BGR2RGB)
mask[mask == 255] = 1
return mask
def img_to_array(img_file):
img = cv2.imread(img_file) # Preprocess image as well and simplify?
img = cv2.cvtColor(img, cv2.COLOR_BGR2RGB)
return img
def fit_model(model, all_training_data):
logger.info('Creating a numpy array')
X = [all_training_data[i][0] for i in range(len(all_training_data))]
Y = [all_training_data[i][1] for i in range(len(all_training_data))]
X = np.array(X)
Y = np.array(Y)
print(X.shape, Y.shape)
for iteration in range(train_iterations):
logger.info('-' * 10 + f'Iteration {iteration+1} out of {train_iterations}' + '-' * 10 )
try:
history = model.fit(X, Y,
epochs=epochs, batch_size=batch_size,
validation_split=0.1, callbacks=[],
shuffle=True, verbose=1)
logger.info('Loss: ' + str(round(history.history['loss'][-1], 4)) + ' Val Loss: '
+ str(round(history.history['val_loss'][-1], 4)))
prediction = model.predict(np.expand_dims(X[0], axis=0))
predicted_mask = predict_to_mask(prediction)
predicted_mask = np.squeeze(predicted_mask, axis=0)
plt.subplot(1,2,1)
plt.title('Real mask')
plt.imshow(predict_to_mask(Y[0]))
plt.subplot(1,2,2)
plt.title('Predicted mask')
plt.imshow(predicted_mask)
if not os.path.isdir('images'):
os.mkdir('images')
plt.savefig(f'images/{iteration}.png')
plt.close()
except KeyboardInterrupt:
model.save('current_weights.h5', overwrite=True)
quit()
def predict_to_mask(mask):
mask = mask*255
return mask
def create_model():
logger.info('Creating model')
model = Sequential()
model.add(Convolution2D(32, 2, 2, input_shape=(1024, 1024, 3), activation='sigmoid'))
model.add(Convolution2D(8, 3, 3, activation='sigmoid'))
model.add(Convolution2D(8, 5, 5, activation='sigmoid'))
model.add(AveragePooling2D(3,3))
model.add(Convolution2D(8, 7, 7, activation='sigmoid'))
model.add(Convolution2D(8, 7, 7, activation='sigmoid'))
model.add(Convolution2D(16, 5, 5, activation='sigmoid'))
model.add(UpSampling2D(2))
for i in range(33):
model.add(Convolution2D(4, 5, 5, activation='sigmoid'))
model.add(Convolution2D(4, 3, 3, activation='relu'))
model.add(AveragePooling2D(2,2))
model.add(UpSampling2D(4))
model.add(Dense(3, activation='sigmoid'))
logger.info(model.summary(print_fn=log_func))
logger.info("Input shape: " + str(model.input_shape))
logger.info("Output shape: " + str(model.output_shape))
return model
"""
logger.info('Creating model')
model = Sequential()
model.add(Convolution2D(128, 20, 20, input_shape=(1024, 1024, 3), activation='relu'))
model.add(Convolution2D(6, 15, 15, activation='sigmoid'))
model.add(Convolution2D(6, 5, 5, activation='sigmoid'))
model.add(AveragePooling2D(3,3))
model.add(Convolution2D(3, 32, 32, activation='sigmoid'))
model.add(Convolution2D(3, 40, 40, activation='sigmoid'))
model.add(Convolution2D(20, 4, 4, activation='sigmoid'))
model.add(UpSampling2D(2))
model.add(AveragePooling2D(2,2))
model.add(UpSampling2D(4))
model.add(Dense(3, activation='sigmoid')) # Softmax since classification
#model.add(Reshape((1024, 1024, 3)))
logger.info(model.summary(print_fn=log_func))
logger.info("Input shape: " + str(model.input_shape))
logger.info("Output shape: " + str(model.output_shape))
return model
"""
def compile_model(model):
logger.info('Compiling model')
adam = Adam(lr=learning_rate) # lr 0.001 --> default adam
sgd = SGD(lr=learning_rate, momentum=0.9, nesterov=True) # lr 0.001 --> default adam
rmsprop = RMSprop(lr=learning_rate)
model.compile(loss='mse', optimizer=adam, metrics = ['binary_accuracy'])
return model
def group_images():
logger.info('Grouping images')
images = glob.glob('Training_dataset/Images/*')
masks = glob.glob('Training_dataset/Masks/all/*')
percentages = glob.glob('Training_dataset/Percentages/*')
all_groups = []
for idx, img in enumerate(images):
group = []
group.append(img)
img_basename = os.path.basename(img).split('.')[0]
for mask in masks:
mask_basename = os.path.basename(mask).split('.')[0]
if mask_basename == img_basename:
group.append(mask)
break
for percent in percentages:
percent_basename = os.path.basename(percent).split('.')[0]
if percent_basename == img_basename:
group.append(percent)
break
if len(group) == 3:
all_groups.append(group)
else:
logger.warning(f'Image {img} not matched with mask and/or percent')
print(f'{idx} done out of {len(images)}', end='\r')
return all_groups
if __name__ == '__main__':
main()