/
train.py
206 lines (167 loc) · 7.84 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
from Resnet_model import resnet101
from utils.data_utils import parser
from utils.plot_utils import ploy_rect
from utils.common_utils import make_summary
import tensorflow as tf
import matplotlib.pyplot as plt
import numpy as np
import random
import math
import sys
import cv2
import csv
import os
# define clasify: background and car
LABELS = {
1: 'Person',
2: 'Bird',
3: 'Cat',
4: 'Cow',
5: 'Dog',
6: 'Horse',
7: 'Sheep',
8: 'Aeroplane',
9: 'Bicycle',
10: 'Boat',
11: 'Bus',
12: 'Car',
13: 'Motorbike',
14: 'Train',
15: 'Bottle',
16: 'Chair',
17: 'Diningtable',
18: 'Pottedplant',
19: 'Sofa',
20: 'TV'
}
TRAIN_TFRECORDS = './tfrecords/voc2012/train*'
VAL_TFRECORDS = './tfrecords/voc2012/val.tfrecords'
# RESTORE_PATH = './yolov3_weight/yolov3.ckpt'
# RESTORE_PATH = './checkpoint/good/RetinaNet.ckpt-80830'
SAVE_DIR = './checkpoint/RetinaNet.ckpt'
TRAIN_NUM = 13600
VAL_NUM = 3400
NUM_CLASS = 20
NUM_ANCHORS = 9
TRAIN_EVAL_INTERNAL = 1360
VAL_EVAL_INTERNAL = 1
SAVE_INTERNAL = 1
EPOCH = 500
BATCH_SIZE = 10
IMAGE_SIZE = [224, 224]
SHUFFLE_SIZE = 500
NUM_PARALLEL = 10
# learning rate and optimizer
OPTIMIZER = 'adam'
LEARNING_RATE_INIT = 1e-4
LEARNING_RATE_TYPE = 'exponential'
LEARNING_RATE_DECAY_STEPS = 300
LEARNING_RATE_DECAY_RATE = 0.96
LEARNING_RATE_MIN = 1e-6
# variables part
RESTORE_PART = ['yolov3/darknet53_body']
UPDATE_PART = ['yolov3/RetinaNet_head']
with tf.Graph().as_default():
train_files = tf.train.match_filenames_once(TRAIN_TFRECORDS)
train_dataset = tf.data.TFRecordDataset(train_files, buffer_size=5)
train_dataset = train_dataset.map(lambda x : parser(x, IMAGE_SIZE, 'train'), num_parallel_calls=NUM_PARALLEL)
train_dataset = train_dataset.shuffle(SHUFFLE_SIZE).batch(BATCH_SIZE).prefetch(BATCH_SIZE)
val_files = [VAL_TFRECORDS]
val_dataset = tf.data.TFRecordDataset(val_files)
val_dataset = val_dataset.map(lambda x : parser(x, IMAGE_SIZE, 'val'), num_parallel_calls=NUM_PARALLEL)
val_dataset = val_dataset.batch(BATCH_SIZE).prefetch(BATCH_SIZE)
# create a public iterator
iterator = tf.data.Iterator.from_structure(train_dataset.output_types,
train_dataset.output_shapes)
image, boxes, labels = iterator.get_next()
image.set_shape([None, IMAGE_SIZE[0], IMAGE_SIZE[1], 3])
with tf.variable_scope("yolov3"):
retinaNet = resnet101(NUM_CLASS, NUM_ANCHORS)
feature_maps, pred_class, pred_boxes = retinaNet.forward(image, is_training=True)
anchors = retinaNet.generate_anchors(feature_maps)
loss, FL_loss, reg_loss = retinaNet.compute_loss(anchors, pred_class, pred_boxes, labels, boxes)
pred_boxes, pred_labels, pred_scores = retinaNet.predict(anchors, pred_class, pred_boxes)
eval_result = retinaNet.evaluate(pred_boxes, pred_labels, pred_scores, boxes, labels)
global_step = tf.Variable(0, trainable=False, collections=[tf.GraphKeys.LOCAL_VARIABLES])
# Define the learning rate and optimizer
learning_rate = tf.train.exponential_decay(LEARNING_RATE_INIT, global_step, LEARNING_RATE_DECAY_STEPS, LEARNING_RATE_DECAY_RATE, staircase=True)
optimizer = tf.train.AdamOptimizer(learning_rate)
# Summary
tf.summary.scalar('loss', loss)
tf.summary.scalar('focal_loss', FL_loss)
tf.summary.scalar('regression_loss', reg_loss)
tf.summary.scalar('learning_rate', learning_rate)
# Restore and update var (if finetuning)
# restore_vars = tf.contrib.framework.get_variables_to_restore(include=RESTORE_PART)
# update_vars = tf.contrib.framework.get_variables_to_restore(include=UPDATE_PART)
# saver_restore = tf.train.Saver(var_list=restore_vars)
# Restore all vars (if continue training)
# saver_restore = tf.train.Saver()
# average model
ema = tf.train.ExponentialMovingAverage(decay=0.99, num_updates=global_step)
ema_op = ema.apply(tf.trainable_variables())
# Update the BN vars
update_ops = tf.get_collection(tf.GraphKeys.UPDATE_OPS)
# train_step = optimizer.minimize(loss, global_step=global_step, var_list=update_vars+restore_vars)
train_step = optimizer.minimize(loss, global_step=global_step)
with tf.control_dependencies(update_ops):
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name='train')
# Set session config
config = tf.ConfigProto(allow_soft_placement=True)
config.gpu_options.per_process_gpu_memory_fraction = 0.80
with tf.Session(config=config) as sess:
sess.run((tf.global_variables_initializer(), tf.local_variables_initializer()))
# saver_restore.restore(sess, RESTORE_PATH)
saver = tf.train.Saver()
write_op = tf.summary.merge_all()
write_train = tf.summary.FileWriter('./log/train', sess.graph)
write_val = tf.summary.FileWriter('./log/val')
print('\n------------- start to train --------------\n')
for epoch in range(EPOCH):
best_precision = 0
sess.run(iterator.make_initializer(train_dataset))
while True:
try:
_, summary_, loss_, FL_loss_, reg_loss_, global_step_, eval_result_ = sess.run([train_op, write_op, loss, FL_loss, reg_loss, global_step, eval_result])
if math.isnan(loss_):
sys.exit(0)
write_train.add_summary(summary_, global_step=global_step_)
info = "EPOCH: {}, global_step: {}, loss: {:.3f}, focal_loss: {:.3f}, reg_loss: {:.3f}".format(epoch, global_step_, loss_, FL_loss_, reg_loss_)
print(info)
# evaluate on the train dataset
if (global_step_ + 1) % TRAIN_EVAL_INTERNAL == 0:
# Calculate the precision and recall
TP_array, pred_label_array, gt_label_array = eval_result_
precision_ = np.sum(TP_array) / (np.sum(pred_label_array) + 1e-6)
recall_ = np.sum(TP_array) / (np.sum(gt_label_array) + 1e-6)
info = "===> batch precision: {:.3f}, batch recall: {:.3f} <===".format(precision_, recall_)
print(info)
except tf.errors.OutOfRangeError:
break
# if (epoch + 1) % SAVE_INTERNAL == 0:
# saver.save(sess, SAVE_DIR, global_step_)
if (epoch + 1) % VAL_EVAL_INTERNAL == 0:
sess.run(iterator.make_initializer(val_dataset))
TP_list, pred_label_list, gt_label_list, loss_list = [], [], [], []
while True:
try:
loss_, eval_result_ = sess.run([loss, eval_result])
TP_array, pred_label_array, gt_label_array = eval_result_
TP_list.append(TP_array)
pred_label_list.append(pred_label_array)
gt_label_list.append(gt_label_array)
loss_list.append(loss_)
except tf.errors.OutOfRangeError:
break
precision = np.sum(np.array(TP_list)) / (np.sum(np.array(pred_label_list)) + 1e-6)
recall = np.sum(np.array(TP_list)) / (np.sum(np.array(gt_label_list)) + 1e-6)
loss_mean = np.mean(np.array(loss_list))
info = "===> Epoch: {}, precision: {:.3f}, recall: {:.3f}".format(epoch, precision, recall)
print(info)
write_val.add_summary(make_summary('val_recall', recall), global_step=epoch)
write_val.add_summary(make_summary('val_precision', precision), global_step=epoch)
# Save the best model
if precision > best_precision:
best_precision = precision
saver.save(sess, SAVE_DIR, epoch)