-
Notifications
You must be signed in to change notification settings - Fork 12
/
Copy pathconfig.py
executable file
·129 lines (107 loc) · 4.47 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
128
from os.path import join as pjoin
from os.path import dirname, abspath
class ConfigTrain(object):
# 目录
PROJECT_ROOT = dirname(abspath(__file__))
DATA_LIST_ROOT = pjoin(PROJECT_ROOT, 'data_list')
TRAIN_ROOT = '/root/data/LaneSeg'
IMAGE_ROOT = pjoin(TRAIN_ROOT, 'Image_Data')
LABEL_ROOT = pjoin(TRAIN_ROOT, 'Gray_Label')
WEIGHTS_ROOT = pjoin(PROJECT_ROOT, 'weights')
WEIGHTS_SAVE_ROOT = pjoin(WEIGHTS_ROOT, '1536x512_b2')
LOG_ROOT = pjoin(PROJECT_ROOT, 'logs')
# log文件
LOG_SUSPICIOUS_FILES = pjoin(LOG_ROOT, 'suspicious_files.log')
# 设备
DEVICE = 'cuda:0'
# 网络类型
NET_NAME = 'unet'
# 网络参数
NUM_CLASSES = 8 # 8个类别
IMAGE_SIZE = (1536, 512) # 训练的图片的尺寸(h,w)
HEIGHT_CROP_OFFSET = 690 # 在height方向上将原图裁掉的offset
BATCH_SIZE = 2 # 数据批次大小
EPOCH_NUM = 30 # 总轮次
PRETRAIN = False # 是否加载预训练的权重
EPOCH_BEGIN = 0 # 接着前面的epoch训练，默认0，表示从头训练
PRETRAINED_WEIGHTS = pjoin(WEIGHTS_ROOT, '1024x384_b4', 'weights_ep_21_0.008_0.013_0.648.pth')
BASE_LR = 0.001 # 学习率
LR_STRATEGY = [
[0.001], # epoch 0
[0.001], # epoch 1
[0.001], # epoch 2
[0.001, 0.0006, 0.0003, 0.0001, 0.0004, 0.0008, 0.001], # epoch 3
[0.001, 0.0006, 0.0003, 0.0001, 0.0004, 0.0008, 0.001], # epoch 4
[0.001, 0.0006, 0.0003, 0.0001, 0.0004, 0.0008, 0.001], # epoch 5
[0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0003, 0.0004], # epoch 6
[0.0004, 0.0003, 0.0002, 0.0001, 0.0002, 0.0003, 0.0004], # epoch 7
[0.0003, 0.0002, 0.0001, 0.0002, 0.0003], # epoch 8
[0.0003, 0.0002, 0.0001, 0.0002, 0.0003], # epoch 9
[0.0003, 0.0002, 0.0001, 0.0002, 0.0003], # epoch 10
[0.0003, 0.0002, 0.0001, 0.0002, 0.0003], # epoch 11
[0.0003, 0.0002, 0.0001, 0.0002, 0.0003], # epoch 12
[0.0002, 0.0001, 0.0002], # epoch 13
[0.0002, 0.0001, 0.0002], # epoch 14
[0.0002, 0.0001, 0.0002], # epoch 15
[0.0002, 0.0001, 0.0002], # epoch 16
[0.0002, 0.0001, 0.0002], # epoch 17
[0.0001], # epoch 18
[0.0001], # epoch 19
[0.0001], # epoch 20
[0.0001], # epoch 21
[0.0001], # epoch 22
[0.0001], # epoch 23
[0.0001], # epoch 24
[0.0001], # epoch 25
[0.0001], # epoch 26
[0.0001], # epoch 27
[0.0001], # epoch 28
[0.0001], # epoch 29
[0.0001], # epoch 30
]
SUSPICIOUS_RATE = 0.75 # 可疑比例：当某个iteration的miou比当前epoch_miou的可疑比例还要小的时候，记录此次iteration的训练数据索引，人工排查是否数据有问题
VAL_BATCH_SIZE = 1 # 验证的数据批次大小
class ConfigInference(object):
# 目录
PROJECT_ROOT = dirname(abspath(__file__))
DATA_ROOT = '/root/data/test'
IMAGE_ROOT = pjoin(DATA_ROOT, 'TestImage')
LABEL_ROOT = pjoin(DATA_ROOT, 'TestLabel')
OVERLAY_ROOT = pjoin(DATA_ROOT, 'TestOverlay')
WEIGHTS_ROOT = pjoin(PROJECT_ROOT, 'weights')
PRETRAINED_WEIGHTS = pjoin(WEIGHTS_ROOT, '1536x512_b2', 'weights_ep_18_0.007_0.011_0.684.pth')
LOG_ROOT = pjoin(PROJECT_ROOT, 'logs')
# 设备
DEVICE = 'cuda:0'
# 网络类型
NET_NAME = 'unet'
# 网络参数
NUM_CLASSES = 8 # 8个类别
IMAGE_SIZE = (1536, 512) # 训练的图片的尺寸(h,w)
HEIGHT_CROP_OFFSET = 690 # 在height方向上将原图裁掉的offset
BATCH_SIZE = 1 # 数据批次大小
# 原图的大小
IMAGE_SIZE_ORG = (3384, 1710)
# 标签模式 color | gray
LABEL_MODE = 'gray'
class ConfigVal(object):
# 目录
PROJECT_ROOT = dirname(abspath(__file__))
DATA_LIST_ROOT = pjoin(PROJECT_ROOT, 'data_list')
VAL_ROOT = '/root/data/LaneSeg'
IMAGE_ROOT = pjoin(VAL_ROOT, 'Image_Data')
LABEL_ROOT = pjoin(VAL_ROOT, 'Gray_Label')
WEIGHTS_ROOT = pjoin(PROJECT_ROOT, 'weights')
VAL_WEIGHTS = pjoin(WEIGHTS_ROOT, '1536x512_b2', 'weights_ep_18_0.007_0.011_0.684.pth')
LOG_ROOT = pjoin(PROJECT_ROOT, 'logs')
# 数据量
VAL_MAX_NUM = 999 # 最多取出多少做训练集
# 设备
DEVICE = 'cuda:0'
# 网络类型
NET_NAME = 'unet'
# 网络参数
NUM_CLASSES = 8 # 8个类别
IMAGE_SIZE = (1536, 512) # 训练的图片的尺寸(h,w)
HEIGHT_CROP_OFFSET = 690 # 在height方向上将原图裁掉的offset
VAL_BATCH_SIZE = 1 # 数据批次大小