-
Notifications
You must be signed in to change notification settings - Fork 99
/
ssd300.py
202 lines (174 loc) · 9.62 KB
/
ssd300.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
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
from tensorflow.keras.layers import Conv2D
from tensorflow.keras.layers import Input
from tensorflow.keras.layers import MaxPooling2D
from tensorflow.keras.layers import ZeroPadding2D
from tensorflow.keras.models import Model
from tensorflow.keras.regularizers import l2
from tensorflow.keras.utils import get_file
from ..layers import Conv2DNormalization
from .utils import create_multibox_head
from .utils import create_prior_boxes
WEIGHT_PATH = (
'https://github.com/oarriaga/altamira-data/releases/download/v0.2/')
def SSD300(num_classes=21, base_weights='VOC', head_weights='VOC',
input_shape=(300, 300, 3), num_priors=[4, 6, 6, 6, 4, 4],
l2_loss=0.0005, return_base=False, trainable_base=True):
"""Single-shot-multibox detector for 300x300x3 BGR input images.
# Arguments
num_classes: Integer. Specifies the number of class labels.
base_weights: String or None. If string should be a valid dataset name.
Current valid datasets include `VOC` `FAT` and `VGG`.
head_weights: String or None. If string should be a valid dataset name.
Current valid datasets include `VOC` and `FAT`.
input_shape: List of integers. Input shape to the model including only
spatial and channel resolution e.g. (300, 300, 3).
num_priors: List of integers. Number of default box shapes
used in each detection layer.
l2_loss: Float. l2 regularization loss for convolutional layers.
return_base: Boolean. If `True` the model returned is just
the original base.
trainable_base: Boolean. If `True` the base model
weights are also trained.
# Reference
- [SSD: Single Shot MultiBox
Detector](https://arxiv.org/abs/1512.02325)
"""
if base_weights not in ['VGG', 'VOC', 'FAT', None]:
raise ValueError('Invalid `base_weights`:', base_weights)
if head_weights not in ['VOC', 'FAT', None]:
raise ValueError('Invalid `base_weights`:', base_weights)
if ((base_weights == 'VGG') and (head_weights is not None)):
raise NotImplementedError('Invalid `base_weights` with head_weights')
if ((base_weights is None) and (head_weights is not None)):
raise NotImplementedError('Invalid `base_weights` with head_weights')
if ((num_classes != 21) and (head_weights == 'VOC')):
raise ValueError('Invalid `head_weights` with given `num_classes`')
if ((num_classes != 22) and (head_weights == 'FAT')):
raise ValueError('Invalid `head_weights` with given `num_classes`')
image = Input(shape=input_shape, name='image')
# Block 1 -----------------------------------------------------------------
conv1_1 = Conv2D(64, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv1_1')(image)
conv1_2 = Conv2D(64, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv1_2')(conv1_1)
pool1 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding='same', )(conv1_2)
# Block 2 -----------------------------------------------------------------
conv2_1 = Conv2D(128, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv2_1')(pool1)
conv2_2 = Conv2D(128, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv2_2')(conv2_1)
pool2 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding='same')(conv2_2)
# Block 3 -----------------------------------------------------------------
conv3_1 = Conv2D(256, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv3_1')(pool2)
conv3_2 = Conv2D(256, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv3_2')(conv3_1)
conv3_3 = Conv2D(256, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv3_3')(conv3_2)
pool3 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding='same')(conv3_3)
# Block 4 -----------------------------------------------------------------
conv4_1 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv4_1')(pool3)
conv4_2 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv4_2')(conv4_1)
conv4_3 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv4_3')(conv4_2)
conv4_3_norm = Conv2DNormalization(20, name='branch_1')(conv4_3)
pool4 = MaxPooling2D(pool_size=(2, 2), strides=(2, 2),
padding='same')(conv4_3)
# Block 5 -----------------------------------------------------------------
conv5_1 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv5_1')(pool4)
conv5_2 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv5_2')(conv5_1)
conv5_3 = Conv2D(512, (3, 3), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='conv5_3')(conv5_2)
pool5 = MaxPooling2D(pool_size=(3, 3), strides=(1, 1),
padding='same')(conv5_3)
# Dense 6/7 --------------------------------------------------------------
pool5z = ZeroPadding2D(padding=(6, 6))(pool5)
fc6 = Conv2D(1024, (3, 3), dilation_rate=(6, 6),
padding='valid', activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='fc6')(pool5z)
fc7 = Conv2D(1024, (1, 1), padding='same',
activation='relu',
kernel_regularizer=l2(l2_loss),
trainable=trainable_base,
name='branch_2')(fc6)
# EXTRA layers in SSD -----------------------------------------------------
# Block 6 -----------------------------------------------------------------
conv6_1 = Conv2D(256, (1, 1), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss))(fc7)
conv6_1z = ZeroPadding2D()(conv6_1)
conv6_2 = Conv2D(512, (3, 3), strides=(2, 2), padding='valid',
activation='relu', name='branch_3',
kernel_regularizer=l2(l2_loss))(conv6_1z)
# Block 7 -----------------------------------------------------------------
conv7_1 = Conv2D(128, (1, 1), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss))(conv6_2)
conv7_1z = ZeroPadding2D()(conv7_1)
conv7_2 = Conv2D(256, (3, 3), padding='valid', strides=(2, 2),
activation='relu', name='branch_4',
kernel_regularizer=l2(l2_loss))(conv7_1z)
# Block 8 -----------------------------------------------------------------
conv8_1 = Conv2D(128, (1, 1), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss))(conv7_2)
conv8_2 = Conv2D(256, (3, 3), padding='valid', strides=(1, 1),
activation='relu', name='branch_5',
kernel_regularizer=l2(l2_loss))(conv8_1)
# Block 9 -----------------------------------------------------------------
conv9_1 = Conv2D(128, (1, 1), padding='same', activation='relu',
kernel_regularizer=l2(l2_loss))(conv8_2)
conv9_2 = Conv2D(256, (3, 3), padding='valid', strides=(1, 1),
activation='relu', name='branch_6',
kernel_regularizer=l2(l2_loss))(conv9_1)
branch_tensors = [conv4_3_norm, fc7, conv6_2, conv7_2, conv8_2, conv9_2]
if return_base:
outputs = branch_tensors
else:
outputs = create_multibox_head(
branch_tensors, num_classes, num_priors, l2_loss)
model = Model(inputs=image, outputs=outputs, name='SSD300')
if ((base_weights is not None) or (head_weights is not None)):
model_filename = ['SSD300', str(base_weights), str(head_weights)]
model_filename = '_'.join(['-'.join(model_filename), 'weights.hdf5'])
weights_path = get_file(model_filename, WEIGHT_PATH + model_filename,
cache_subdir='paz/models')
print('Loading %s model weights' % weights_path)
finetunning_model_names = ['SSD300-VGG-None_weights.hdf5',
'SSD300-VOC-None_weights.hdf5']
by_name = True if model_filename in finetunning_model_names else False
model.load_weights(weights_path, by_name=by_name)
model.prior_boxes = create_prior_boxes('VOC')
return model