-
Notifications
You must be signed in to change notification settings - Fork 0
/
yolo.py
139 lines (102 loc) · 4.82 KB
/
yolo.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
import tensorflow as tf
import backbone
import common_layer as cl
from tensorflow.keras import layers
from tensorflow import keras
def yolo_v3(input_layer, NUM_CLASS):
route_1, route_2, conv = backbone.darknet53(input_layer)
# 1st detector
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv = cl.convolution(conv, (3, 3, 512, 1024))
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv = cl.convolution(conv, (3, 3, 512, 1024))
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv_lobj_branch = cl.convolution(conv, (3, 3, 512, 1024))
conv_lbbox = cl.convolution(conv_lobj_branch, (1, 1, 1024, 3 * (1 + 4 + NUM_CLASS)), activate=False, bn=False)
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.upsample(conv)
conv = tf.concat([conv, route_2], axis=-1)
# 2nd detector
conv = cl.convolution(conv, (1, 1, 768, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
conv_mobj_branch = cl.convolution(conv, (3, 3, 256, 512))
conv_mbbox = cl.convolution(conv_mobj_branch, (1, 1, 512, 3 * (1 + 4 + NUM_CLASS)), activate=False, bn=False)
conv = cl.convolution(conv, (1, 1, 256, 128))
conv = cl.upsample(conv)
conv = tf.concat([conv, route_1], axis=-1)
# 3rd detector
conv = cl.convolution(conv, (1, 1, 384, 128))
conv = cl.convolution(conv, (3, 3, 128, 256))
conv = cl.convolution(conv, (1, 1, 256, 128))
conv = cl.convolution(conv, (3, 3, 128, 256))
conv = cl.convolution(conv, (1, 1, 256, 128))
conv_sobj_branch = cl.convolution(conv, (3, 3, 128, 256))
conv_sbbox = cl.convolution(conv_sobj_branch, (1, 1, 256, 3 * (1 + 4 + NUM_CLASS)), activate=False, bn=False)
return [conv_sbbox, conv_mbbox, conv_lbbox]
def yolo_v4(input_layer, NUM_CLASS):
route_1, route_2, conv = backbone.cspdarknet53(input_layer)
route = conv
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.upsample(conv)
route_2 = cl.convolution(route_2, (1, 1, 512, 256))
conv = tf.concat([route_2, conv], axis=-1)
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
route_2 = conv
conv = cl.convolution(conv, (1, 1, 256, 128))
conv = cl.upsample(conv)
route_1 = cl.convolution(route_1, (1, 1, 256, 128))
conv = tf.concat([route_1, conv], axis=-1)
conv = cl.convolution(conv, (1, 1, 256, 128))
conv = cl.convolution(conv, (3, 3, 128, 256))
conv = cl.convolution(conv, (1, 1, 256, 128))
conv = cl.convolution(conv, (3, 3, 128, 256))
conv = cl.convolution(conv, (1, 1, 256, 128))
route_1 = conv
conv = cl.convolution(conv, (3, 3, 128, 256))
conv_sbbox = cl.convolution(conv, (1, 1, 256, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
conv = cl.convolution(route_1, (3, 3, 128, 256), downsample=True)
conv = tf.concat([conv, route_2], axis=-1)
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
conv = cl.convolution(conv, (3, 3, 256, 512))
conv = cl.convolution(conv, (1, 1, 512, 256))
route_2 = conv
conv = cl.convolution(conv, (3, 3, 256, 512))
conv_mbbox = cl.convolution(conv, (1, 1, 512, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
conv = cl.convolution(route_2, (3, 3, 256, 512), downsample=True)
conv = tf.concat([conv, route], axis=-1)
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv = cl.convolution(conv, (3, 3, 512, 1024))
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv = cl.convolution(conv, (3, 3, 512, 1024))
conv = cl.convolution(conv, (1, 1, 1024, 512))
conv = cl.convolution(conv, (3, 3, 512, 1024))
conv_lbbox = cl.convolution(conv, (1, 1, 1024, 3 * (NUM_CLASS + 5)), activate=False, bn=False)
return [conv_sbbox, conv_mbbox, conv_lbbox]
def yolo_v4_TF():
x_input = tf.keras.Input(shape=(608, 608, 3))
x = layers.Conv2D(filters=32,
kernel_size=3,
strides=1,
use_bias=False,
padding='same',
kernel_regularizer=tf.keras.regularizers.l2(0.0005))(x_input)
x = layers.BatchNormalization()(x)
x = x * tf.math.tanh(tf.math.softplus(x))
x = layers.Conv2D(filters=64,
kernel_size=3,
strides=2)(x)
x = x * tf.math.tanh(tf.math.softplus(x))
x_output = layers.Dense(4)(x)
model = tf.keras.Model(x_input, x_output)
model.summary()
tf.keras.utils.plot_model(model, "yolo_v4_TF.png", show_shapes=True)
yolo_v4_TF()