-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathtrainer.py
121 lines (93 loc) · 4.4 KB
/
trainer.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
from matplotlib import pyplot as plt
import numpy as np
import tensorflow as tf
import sys
from loader.DataLoader import load
from models.net import network
import lib.loss as losses
from config.parameters import Config
import lib.utils as utils
import lib.ls as lss
C = Config()
def get_img_output_length(width, height):
return (int(width/16),int(height/16))
roi_input = tf.placeholder(tf.float32, shape=[1, None, 4])
num_rois = 4
num_epo = 500
dataset_path = sys.argv[1]
load = load(dataset_path)
data = load.get_data()
num_anchors = 9
data_gen = load.get_anchor_gt(data, C, get_img_output_length, mode='train')
net = network()
initializer = tf.random_normal_initializer(mean=0.0, stddev=0.01)
initializer_bbox = tf.random_normal_initializer(mean=0.0, stddev=0.001)
import pdb; pdb.set_trace()
rpn_out = net.build_network()
x, cls_plc, box_plc = net.getPlaceholders()
lsr = lss.rpn_loss_cls_org(9)
lgr = lss.rpn_loss_regr_org(9)
los_c = lsr(cls_plc, rpn_out[0])
los_b = lgr(box_plc, rpn_out[1])
rpn_loss = los_c + los_b
# rpn_loss = losses.rpn()
class_mapping = {'human':0, 'raccoon':1}
classifier = net.classifier(rpn_out[2], roi_input, num_rois, nb_classes=len(class_mapping), trainable=True)
lab_cls = tf.placeholder(tf.float32, shape=[1, None, 1], name='label_class')
lab_reg = tf.placeholder(tf.float32, shape=[1, None, 4], name='label_regression')
clf = lss.class_loss_regr(1)
clf_cls = lss.class_loss_cls(lab_cls, classifier[0])
clf_reg = clf(lab_reg, classifier[1])
clf_loss = clf_cls + clf_reg
total_loss = rpn_loss + clf_loss
tf.summary.scalar("loss", total_loss)
train_step = tf.train.AdamOptimizer(1e-4).minimize(total_loss)
saver = tf.train.Saver()
with tf.Session() as sess:
train_writer = tf.summary.FileWriter( 'logs/', sess.graph)
merged = tf.summary.merge_all()
sess.run(tf.global_variables_initializer())
for i in range(num_epo):
los = 0
for _ in range(256):
X, Y, image_data, debug_img, debug_num_pos = next(data_gen)
P_rpn = sess.run(rpn_out, feed_dict={x:X, cls_plc:Y[0], box_plc:Y[1]})
R = utils.rpn_to_roi(P_rpn[0], P_rpn[1], C, 'tf', use_regr=True, overlap_thresh=0.7, max_boxes=300)
X2, Y1, Y2, IouS = utils.calc_iou(R, image_data, C, class_mapping)
neg_samples = np.where(Y1[0, :, -1] == 1)
pos_samples = np.where(Y1[0, :, -1] == 0)
if len(neg_samples) > 0:
neg_samples = neg_samples[0]
else:
neg_samples = []
if len(pos_samples) > 0:
pos_samples = pos_samples[0]
else:
pos_samples = []
if num_rois > 1:
if len(pos_samples) < num_rois//2:
selected_pos_samples = pos_samples.tolist()
else:
selected_pos_samples = np.random.choice(pos_samples, num_rois//2, replace=False).tolist()
try:
selected_neg_samples = np.random.choice(neg_samples, num_rois - len(selected_pos_samples), replace=False).tolist()
except:
selected_neg_samples = np.random.choice(neg_samples, num_rois - len(selected_pos_samples), replace=True).tolist()
sel_samples = selected_pos_samples + selected_neg_samples
else:
# in the extreme case where num_rois = 1, we pick a random pos or neg sample
selected_pos_samples = pos_samples.tolist()
selected_neg_samples = neg_samples.tolist()
if np.random.randint(0, 2):
sel_samples = random.choice(neg_samples)
else:
sel_samples = random.choice(pos_samples)
summary = sess.run([merged, train_step], feed_dict={rpn_out[2]:P_rpn[2], roi_input:X2[:, sel_samples, :], lab_cls:Y1[:, sel_samples, :1], lab_reg:Y2[:, sel_samples, :], x:X, cls_plc:Y[0], box_plc:Y[1]})
ls_val = sess.run(total_loss, feed_dict={rpn_out[2]:P_rpn[2], roi_input:X2[:, sel_samples, :], lab_cls:Y1[:, sel_samples, :1], lab_reg:Y2[:, sel_samples, :], x:X, cls_plc:Y[0], box_plc:Y[1]})
loss_ = ls_val + los
los = loss_
train_writer.add_summary(summary[0], i)
print ("epoch : %s ***** losss : %s ***** "%(i, loss_/256))
if i%100 == 0:
save_path = saver.save(sess, 'weight/'+"model_{}.ckpt".format(i))
print ("epoch : %s saved at %s "%(i, save_path))