-
Notifications
You must be signed in to change notification settings - Fork 103
/
Copy pathconfig.ini
executable file
·137 lines (117 loc) · 2.97 KB
/
config.ini
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
[config]
root = ~/model/yolo2-pytorch
[image]
size = 416 416
[cache]
name = cache
category = config/category/20
# voc coco
datasets = cache.voc.cache cache.coco.cache
shuffle = 1
[model]
name = model
anchors = config/anchors/voc.tsv
; model.yolo2.Darknet
; model.yolo2.Tiny
; model.resnet.resnet18
; model.inception3.Inception3
; model.inception4.Inception4
; model.mobilenet.MobileNet
; model.densenet.densenet121
; model.vgg.vgg19
dnn = model.yolo2.Tiny
pretrained = 0
threshold = 0.6
[batch_norm]
enable = 1
gamma = 1
beta = 1
[inception4]
pretrained = imagenet
[data]
workers = 3
sizes = 320,320 352,352 384,384 416,416 448,448 480,480 512,512 544,544 576,576 608,608
maintain = 10
shuffle = 0
# rescale padding
resize = rescale
[transform]
; transform.augmentation.RandomRotate transform.augmentation.RandomFlipHorizontally
augmentation = transform.augmentation.RandomRotate transform.augmentation.RandomFlipHorizontally
resize_train = transform.resize.label.RandomCrop
resize_eval = transform.resize.label.Resize
resize_test = transform.resize.image.Resize
; transform.image.RandomBlur transform.image.BGR2HSV transform.image.RandomHue transform.image.RandomSaturation transform.image.RandomBrightness transform.image.HSV2RGB transform.image.RandomGamma
image_train = transform.image.BGR2RGB
image_test = transform.image.BGR2RGB
; torchvision.transforms.ToTensor transform.image.Normalize
tensor = torchvision.transforms.ToTensor transform.image.Normalize
normalize = 0.5 1
[augmentation]
random_rotate = -5 5
random_flip_horizontally = 0.5
random_crop = 1
random_blur = 5 5
random_hue = 0 25
random_saturation = 0.5 1.5
random_brightness = 0.5 1.5
random_gamma = 0.9 1.5
[train]
; lambda params, lr: torch.optim.SGD(params, lr, momentum=2)
; lambda params, lr: torch.optim.Adam(params, lr, betas=(0.9, 0.999), eps=1e-8)
; lambda params, lr: torch.optim.RMSprop(params, lr, alpha=0.99, eps=1e-8)
optimizer = lambda params, lr: torch.optim.Adam(params, lr, betas=(0.9, 0.999), eps=1e-8)
; lambda optimizer: torch.optim.lr_scheduler.StepLR(optimizer, step_size=30, gamma=0.1)
; lambda optimizer: torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[60, 90], gamma=0.1)
scheduler = lambda optimizer: torch.optim.lr_scheduler.MultiStepLR(optimizer, milestones=[60, 90], gamma=0.1)
phase = train val
cross_entropy = 1
clip_ = 5
[save]
secs = 600
keep = 5
[summary]
scalar = 10
image = 60
histogram_ = 60
[summary_scalar]
loss_hparam = 1
[summary_image]
limit = 2
bbox = 1
iou = 1
[summary_histogram]
parameters = config/summary/histogram.txt
[hparam]
foreground = 5
background = 1
center = 1
size = 1
cls = 1
[detect]
threshold = 0.3
threshold_cls = 0.005
fix = 0
overlap = 0.45
[eval]
phase = test
secs = 12 * 60 * 60
first = 0
iou = 0.5
db = eval.json
mapper = config/eval.py
debug = 0
sort = timestamp
metric07 = 1
[graph]
metric = lambda t: np.mean(utils.dense(t))
format = svg
[digraph_graph_attr]
size = 12, 12
[digraph_node_attr]
style = filled
shape = box
align = left
fontsize = 12
ranksep = 0.1
height = 0.2