|
a |
|
b/select_model.py |
|
|
1 |
from Segmentation.model.unet import UNet, R2_UNet, Nested_UNet, Nested_UNet_v2 |
|
|
2 |
from Segmentation.model.segnet import SegNet |
|
|
3 |
from Segmentation.model.deeplabv3 import Deeplabv3, Deeplabv3_plus |
|
|
4 |
from Segmentation.model.vnet import VNet |
|
|
5 |
from Segmentation.model.Hundred_Layer_Tiramisu import Hundred_Layer_Tiramisu |
|
|
6 |
|
|
|
7 |
from absl import logging |
|
|
8 |
|
|
|
9 |
def select_model(FLAGS, num_classes): |
|
|
10 |
|
|
|
11 |
if FLAGS.model_architecture == 'unet': |
|
|
12 |
model_args = [FLAGS.num_filters, |
|
|
13 |
num_classes, |
|
|
14 |
FLAGS.use_2d, |
|
|
15 |
FLAGS.backbone_architecture, |
|
|
16 |
FLAGS.num_conv, |
|
|
17 |
FLAGS.kernel_size, |
|
|
18 |
FLAGS.activation, |
|
|
19 |
FLAGS.use_attention, |
|
|
20 |
FLAGS.use_batchnorm, |
|
|
21 |
FLAGS.use_bias, |
|
|
22 |
FLAGS.use_dropout, |
|
|
23 |
FLAGS.dropout_rate, |
|
|
24 |
FLAGS.use_spatial, |
|
|
25 |
FLAGS.channel_order] |
|
|
26 |
|
|
|
27 |
model_fn = UNet |
|
|
28 |
elif FLAGS.model_architecture == 'vnet': |
|
|
29 |
model_args = [FLAGS.num_filters, |
|
|
30 |
num_classes, |
|
|
31 |
FLAGS.use_2d, |
|
|
32 |
FLAGS.num_conv, |
|
|
33 |
FLAGS.kernel_size, |
|
|
34 |
FLAGS.activation, |
|
|
35 |
FLAGS.use_batchnorm, |
|
|
36 |
FLAGS.dropout_rate, |
|
|
37 |
FLAGS.use_spatial, |
|
|
38 |
FLAGS.channel_order] |
|
|
39 |
model_fn = VNet |
|
|
40 |
elif FLAGS.model_architecture == 'r2unet': |
|
|
41 |
model_args = [FLAGS.num_filters, |
|
|
42 |
num_classes, |
|
|
43 |
FLAGS.use_2d, |
|
|
44 |
FLAGS.num_conv, |
|
|
45 |
FLAGS.kernel_size, |
|
|
46 |
FLAGS.activation, |
|
|
47 |
2, |
|
|
48 |
FLAGS.use_attention, |
|
|
49 |
FLAGS.use_batchnorm, |
|
|
50 |
FLAGS.use_bias, |
|
|
51 |
FLAGS.channel_order] |
|
|
52 |
model_fn = R2_UNet |
|
|
53 |
|
|
|
54 |
elif FLAGS.model_architecture == 'segnet': |
|
|
55 |
model_args = [FLAGS.num_filters, |
|
|
56 |
num_classes, |
|
|
57 |
FLAGS.backbone_architecture, |
|
|
58 |
FLAGS.kernel_size, |
|
|
59 |
(2, 2), |
|
|
60 |
FLAGS.activation, |
|
|
61 |
FLAGS.use_batchnorm, |
|
|
62 |
FLAGS.use_bias, |
|
|
63 |
FLAGS.use_transpose, |
|
|
64 |
FLAGS.use_dropout, |
|
|
65 |
FLAGS.dropout_rate, |
|
|
66 |
FLAGS.use_spatial, |
|
|
67 |
FLAGS.channel_order] |
|
|
68 |
|
|
|
69 |
model_fn = SegNet |
|
|
70 |
|
|
|
71 |
elif FLAGS.model_architecture == 'unet++': |
|
|
72 |
model_args = [FLAGS.num_filters, |
|
|
73 |
num_classes, |
|
|
74 |
FLAGS.num_conv, |
|
|
75 |
FLAGS.kernel_size, |
|
|
76 |
FLAGS.activation, |
|
|
77 |
FLAGS.use_batchnorm, |
|
|
78 |
FLAGS.use_bias, |
|
|
79 |
FLAGS.channel_order] |
|
|
80 |
model_fn = Nested_UNet |
|
|
81 |
|
|
|
82 |
elif FLAGS.model_architecture == '100-Layer-Tiramisu': |
|
|
83 |
model_args = [FLAGS.growth_rate, |
|
|
84 |
FLAGS.layers_per_block, |
|
|
85 |
FLAGS.init_num_channels, |
|
|
86 |
num_classes, |
|
|
87 |
FLAGS.kernel_size, |
|
|
88 |
FLAGS.pool_size, |
|
|
89 |
FLAGS.activation, |
|
|
90 |
FLAGS.dropout_rate, |
|
|
91 |
FLAGS.strides, |
|
|
92 |
FLAGS.padding] |
|
|
93 |
|
|
|
94 |
model_fn = Hundred_Layer_Tiramisu |
|
|
95 |
|
|
|
96 |
elif FLAGS.model_architecture == 'deeplabv3': |
|
|
97 |
model_args = [num_classes, |
|
|
98 |
FLAGS.kernel_size_initial_conv, |
|
|
99 |
FLAGS.num_filters_atrous, |
|
|
100 |
FLAGS.num_filters_DCNN, |
|
|
101 |
FLAGS.num_filters_ASPP, |
|
|
102 |
FLAGS.kernel_size_atrous, |
|
|
103 |
FLAGS.kernel_size_DCNN, |
|
|
104 |
FLAGS.kernel_size_ASPP, |
|
|
105 |
'same', |
|
|
106 |
FLAGS.activation, |
|
|
107 |
FLAGS.use_batchnorm, |
|
|
108 |
FLAGS.use_bias, |
|
|
109 |
FLAGS.channel_order, |
|
|
110 |
FLAGS.MultiGrid, |
|
|
111 |
FLAGS.rate_ASPP, |
|
|
112 |
FLAGS.output_stride] |
|
|
113 |
|
|
|
114 |
model_fn = Deeplabv3 |
|
|
115 |
|
|
|
116 |
elif FLAGS.model_architecture == 'deeplabv3_plus': |
|
|
117 |
model_args = [num_classes, |
|
|
118 |
FLAGS.kernel_size_initial_conv, |
|
|
119 |
FLAGS.num_filters_atrous, |
|
|
120 |
FLAGS.num_filters_DCNN, |
|
|
121 |
FLAGS.num_filters_ASPP, |
|
|
122 |
FLAGS.kernel_size_atrous, |
|
|
123 |
FLAGS.kernel_size_DCNN, |
|
|
124 |
FLAGS.kernel_size_ASPP, |
|
|
125 |
FLAGS.num_filters_final_encoder, |
|
|
126 |
FLAGS.num_filters_from_backbone, |
|
|
127 |
FLAGS.num_channels_UpConv, |
|
|
128 |
FLAGS.kernel_size_UpConv, |
|
|
129 |
(2, 2), |
|
|
130 |
False, |
|
|
131 |
FLAGS.use_transpose, |
|
|
132 |
'same', |
|
|
133 |
FLAGS.activation, |
|
|
134 |
FLAGS.use_batchnorm, |
|
|
135 |
FLAGS.use_bias, |
|
|
136 |
FLAGS.channel_order, |
|
|
137 |
FLAGS.MultiGrid, |
|
|
138 |
FLAGS.rate_ASPP, |
|
|
139 |
FLAGS.output_stride] |
|
|
140 |
|
|
|
141 |
model_fn = Deeplabv3_plus |
|
|
142 |
|
|
|
143 |
else: |
|
|
144 |
logging.error('The model architecture {} is not supported!'.format(FLAGS.model_architecture)) |