-
Notifications
You must be signed in to change notification settings - Fork 0
/
config.py
127 lines (104 loc) · 5.5 KB
/
config.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
import os
"""preprocess"""
# 数据集根路径
current_working_directory = os.getcwd()
data_base_dir = os.path.join(current_working_directory, 'data')
# pnet, rnet, onet的数据目录
pnet_data_dir = os.path.join(data_base_dir, 'pnet_data')
rnet_data_dir = os.path.join(data_base_dir, 'rnet_data')
onet_data_dir = os.path.join(data_base_dir, 'onet_data')
# pnet的各种图片及其img_list文本文件的路径
pnet_pos_dir = os.path.join(pnet_data_dir, 'positive')
pnet_neg_dir = os.path.join(pnet_data_dir, 'negative')
pnet_part_dir = os.path.join(pnet_data_dir, 'part')
pnet_landmark_dir = os.path.join(pnet_data_dir, 'landmark')
pnet_pos_img_list = os.path.join(pnet_data_dir, 'positive_img_list.txt')
pnet_neg_img_list = os.path.join(pnet_data_dir, 'negative_img_list.txt')
pnet_part_img_list = os.path.join(pnet_data_dir, 'part_img_list.txt')
pnet_landmark_img_list = os.path.join(pnet_data_dir, 'landmark_img_list.txt')
pnet_mixed_img_list = os.path.join(pnet_data_dir, 'mixed_img_list.txt')
pnet_tfrecord_dir = os.path.join(pnet_data_dir, 'tfrecord')
pnet_tfrecord_file = os.path.join(pnet_tfrecord_dir, 'tfrecord')
# rnet的各种图片及其img_list文本文件的路径
rnet_pos_dir = os.path.join(rnet_data_dir, 'positive')
rnet_neg_dir = os.path.join(rnet_data_dir, 'negative')
rnet_part_dir = os.path.join(rnet_data_dir, 'part')
rnet_landmark_dir = os.path.join(rnet_data_dir, 'landmark')
rnet_pos_img_list = os.path.join(rnet_data_dir, 'positive_img_list.txt')
rnet_neg_img_list = os.path.join(rnet_data_dir, 'negative_img_list.txt')
rnet_part_img_list = os.path.join(rnet_data_dir, 'part_img_list.txt')
rnet_landmark_img_list = os.path.join(rnet_data_dir, 'landmark_img_list.txt')
rnet_tfrecord_dir = os.path.join(rnet_data_dir, 'tfrecord')
rnet_tfrecord_file_pos = os.path.join(rnet_tfrecord_dir, 'pos_img_list_tfrecord')
rnet_tfrecord_file_neg = os.path.join(rnet_tfrecord_dir, 'neg_img_list_tfrecord')
rnet_tfrecord_file_part = os.path.join(rnet_tfrecord_dir, 'part_img_list_tfrecord')
rnet_tfrecord_file_landmark = os.path.join(rnet_tfrecord_dir, 'landmark_img_list_tfrecord')
# onet的各种图片及其img_list文本文件的路径
onet_pos_dir = os.path.join(onet_data_dir, 'positive')
onet_neg_dir = os.path.join(onet_data_dir, 'negative')
onet_part_dir = os.path.join(onet_data_dir, 'part')
onet_landmark_dir = os.path.join(onet_data_dir, 'landmark')
onet_pos_img_list = os.path.join(onet_data_dir, 'positive_img_list.txt')
onet_neg_img_list = os.path.join(onet_data_dir, 'negative_img_list.txt')
onet_part_img_list = os.path.join(onet_data_dir, 'part_img_list.txt')
onet_landmark_img_list = os.path.join(onet_data_dir, 'landmark_img_list.txt')
onet_tfrecord_dir = os.path.join(onet_data_dir, 'tfrecord')
onet_tfrecord_file_pos = os.path.join(onet_tfrecord_dir, 'pos_img_list_tfrecord')
onet_tfrecord_file_neg = os.path.join(onet_tfrecord_dir, 'neg_img_list_tfrecord')
onet_tfrecord_file_part = os.path.join(onet_tfrecord_dir, 'part_img_list_tfrecord')
onet_tfrecord_file_landmark = os.path.join(onet_tfrecord_dir, 'landmark_img_list_tfrecord')
# WIDER数据集路径，WIDER图像的路径及其label
WIDER_base_dir = os.path.join(data_base_dir, 'WIDER')
WIDER_train_dir = os.path.join(WIDER_base_dir, 'WIDER_train')
WIDER_train_images_dir = os.path.join(WIDER_train_dir, 'images')
WIDER_labels_dir = os.path.join(WIDER_base_dir, 'wider_face_split')
wider_face_train_label = os.path.join(WIDER_labels_dir, 'wider_face_train.txt')
wider_face_train_bbx_gt = os.path.join(WIDER_labels_dir, 'wider_face_train_bbx_gt.txt')
# landmark数据集（lfw）的label
LFW_base_dir = os.path.join(data_base_dir, 'LFW')
lfw_train_labels = os.path.join(LFW_base_dir, 'trainImageList.txt')
""" train """
# 批次大小
batch_size = 384
# 三个网络的阈值
net_thresholds = [0.7, 0.8, 0.96]
# ohem策略中使用的有效样本比例
ohem_keep_ratio = 0.8
# 训练日志保存的位置
logs_dir = os.path.join(data_base_dir, 'logs')
train_pnet_log_file = os.path.join(logs_dir, 'train_pnet.log')
train_rnet_log_file = os.path.join(logs_dir, 'train_rnet.log')
train_onet_log_file = os.path.join(logs_dir, 'train_onet.log')
# 日志文件的解析输出
train_pnet_metrics_file = os.path.join(logs_dir, 'train_pnet_metrics')
train_rnet_metrics_file = os.path.join(logs_dir, 'train_rnet_metrics')
train_onet_metrics_file = os.path.join(logs_dir, 'train_onet_metrics')
# checkpoint保存位置
base_checkpoint_dir = 'model_checkpoints'
pnet_checkpoint_dir = os.path.join(data_base_dir, base_checkpoint_dir, 'pnet')
rnet_checkpoint_dir = os.path.join(data_base_dir, base_checkpoint_dir, 'rnet')
onet_checkpoint_dir = os.path.join(data_base_dir, base_checkpoint_dir, 'onet')
checkpoint_dirs = [pnet_checkpoint_dir, rnet_checkpoint_dir, onet_checkpoint_dir]
# 不同损失的权重
classification_loss_weight = 5
bbox_loss_weight = 2
landmark_loss_weight = 2
# 迭代次数
num_epochs = [60, 50, 40]
# 初始学习率
lr = 0.001
# 学习率减少的迭代次数
decay_epochs = [6, 16, 24]
# 经过多少epoch后显示log
log_step = 100
"""test"""
# 测试集所在路径
test_base_dir = os.path.join(data_base_dir, 'test')
# 测试图片放置位置
input_imgs_dir = os.path.join(test_base_dir, 'input_imgs')
# 测试图片输出位置
output_imgs_dir = os.path.join(test_base_dir, 'output_imgs')
# 测试视频放置位置
input_videos_dir = os.path.join(test_base_dir, 'input_videos')
# 测试视频输出位置
output_videos_dir = os.path.join(test_base_dir, 'output_videos')