没有在网上查到tensorflow2实现的vnet网络结构，自己写了一个，供自己和大家参考。

 

import warnings
warnings.filterwarnings("ignore")
import os
os.environ['TF_CPP_MIN_LOG_LEVEL'] = '3'from tensorflow.keras.layers import (Conv3D,PReLU,Input,Concatenate,Conv3DTranspose,)
from tensorflow.keras.models import Modeldef GetVnet(num_classes=2):inputs = Input(shape=(128,128,64,1),name='vnet_input')conv1 = Conv3D(16,5,padding='same',name='conv1')(inputs)conv1 = PReLU(name='conv1_prelu')(conv1) # activationconcat1 = Concatenate(name='concat1')([inputs,conv1])down_conv1 = Conv3D(32,2,2,name='down_conv_1')(concat1)down_conv1 = PReLU(name='down_conv1_prelu')(down_conv1)conv2_1 = Conv3D(32,5,padding='same',name='conv2_1')(down_conv1)conv2_1 = PReLU(name='conv2_1_prelu')(conv2_1)conv2_2 = Conv3D(32,5,padding='same',name='conv2_2')(conv2_1)conv2_2 = PReLU(name='conv2_2_prelu')(conv2_2)concat2 = Concatenate(name='concat2')([down_conv1,conv2_2])down_conv2 = Conv3D(64,2,2,name='down_conv2')(concat2)down_conv2 = PReLU(name='down_conv2_prelu')(down_conv2)conv3_1 = Conv3D(64,5,padding='same',name='conv3_1')(down_conv2)conv3_1 = PReLU(name='conv3_1_prelu')(conv3_1)conv3_2 = Conv3D(64,5,padding='same',name='conv3_2')(conv3_1)conv3_2 = PReLU(name='conv3_2_prelu')(conv3_2)conv3_3 = Conv3D(64,5,padding='same',name='conv3_3')(conv3_2)conv3_3 = PReLU(name='conv3_3_prelu')(conv3_3)concat3 = Concatenate(name='concat3')([conv3_3,down_conv2])down_conv3 = Conv3D(128,2,2,name='down_conv3')(concat3)down_conv3 = PReLU(name='down_conv3_prelu')(down_conv3)conv4_1 = Conv3D(128,5,padding='same',name='conv4_1')(down_conv3)conv4_1 = PReLU(name='conv4_1_prelu')(conv4_1)conv4_2 = Conv3D(128,5,padding='same',name='conv4_2')(conv4_1)conv4_2 = PReLU(name='conv4_2_prelu')(conv4_2)conv4_3 = Conv3D(128,5,padding='same',name='conv4_3')(conv4_2)conv4_3 = PReLU(name='conv4_3_prelu')(conv4_3)concat4 = Concatenate(name='concat4')([down_conv3,conv4_3])down_conv4 = Conv3D(256,2,2,name='down_conv4')(concat4)down_conv4 = PReLU(name='down_conv4_prelu')(down_conv4)conv5_1 = Conv3D(256,5,padding='same',name='conv5_1')(down_conv4)conv5_1 = PReLU(name='conv5_1_prelu')(conv5_1)conv5_2 = Conv3D(256,5,padding='same',name='conv5_2')(conv5_1)conv5_2 = PReLU(name='conv5_2_prelu')(conv5_2)conv5_3 = Conv3D(256,5,padding='same',name='conv5_3')(conv5_2)conv5_3 = PReLU(name='conv5_3_prelu')(conv5_3)concat5 = Concatenate(name='concat5')([down_conv4,conv5_3])up_conv5 = Conv3DTranspose(256,2,2,name='up_conv5')(concat5)up_conv5 = PReLU(name='up_conv5_prelu')(up_conv5)conv6 = Concatenate(name='conv6')([up_conv5,concat4])conv6_1 = Conv3D(256,5,padding='same',name='conv6_1')(conv6)conv6_1 = PReLU(name='conv6_1_prelu')(conv6_1)conv6_2 = Conv3D(256,5,padding='same',name='conv6_2')(conv6_1)conv6_2 = PReLU(name='conv6_2_prelu')(conv6_2)conv6_3 = Conv3D(256,5,padding='same',name='conv6_3')(conv6_2)conv6_3 = PReLU(name='conv6_3_prelu')(conv6_3)concat6 = Concatenate(name='concat6')([up_conv5,conv6_3])up_conv6 = Conv3DTranspose(128,2,2,name='up_conv6')(concat6)up_conv6 = PReLU(name='up_conv6_prelu')(up_conv6)conv7 = Concatenate(name='conv7')([up_conv6,concat3])""" 3次卷积 """for i in range(1,4):conv7 = Conv3D(128,5,padding='same',name=f'conv7_{i}')(conv7)conv7 = PReLU(name=f'conv7_{i}_prelu')(conv7)concat7 = Concatenate(name='concat7')([up_conv6,conv7])up_conv7 = Conv3DTranspose(64,2,2,name='up_conv7')(concat7)up_conv7 = PReLU(name='up_conv7_prelu')(up_conv7)conv8 = Concatenate(name='conv8')([up_conv7,concat2])""" 2次卷积 """for i in range(1,3):conv8 = Conv3D(64,5,padding='same',name=f'conv8_{i}')(conv8)conv8 = PReLU(name=f'conv8_{i}_prelu')(conv8)concat8 = Concatenate(name='concat8')([conv8,up_conv7])up_conv8 = Conv3DTranspose(32,2,2,name='up_conv8')(concat8)up_conv8 = PReLU(name='up_conv8_prelu')(up_conv8)conv9 = Concatenate(name='conv9')([up_conv8,concat1])conv9_1 = Conv3D(32,5,padding='same',name='conv9_1')(conv9)concat9 = Concatenate(name='concat9')([conv9_1,up_conv8])conv10 = Conv3D(num_classes,1,1,name='out',activation='softmax')(concat9)model = Model(inputs=inputs,outputs=conv10,name='myTf2_vnet')return modelif __name__ == '__main__':model = GetVnet()model.summary()

输出：

D:\program\miniconda\envs\py38_tf2\python.exe E:/paper/nets/vnet/Vnet.py
Model: "myTf2_vnet"
__________________________________________________________________________________________________Layer (type)                   Output Shape         Param #     Connected to                     
==================================================================================================vnet_input (InputLayer)        [(None, 128, 128, 6  0           []                               4, 1)]                                                            conv1 (Conv3D)                 (None, 128, 128, 64  2016        ['vnet_input[0][0]']             , 16)                                                             conv1_prelu (PReLU)            (None, 128, 128, 64  16777216    ['conv1[0][0]']                  , 16)                                                             concat1 (Concatenate)          (None, 128, 128, 64  0           ['vnet_input[0][0]',             , 17)                             'conv1_prelu[0][0]']            down_conv_1 (Conv3D)           (None, 64, 64, 32,   4384        ['concat1[0][0]']                32)                                                               down_conv1_prelu (PReLU)       (None, 64, 64, 32,   4194304     ['down_conv_1[0][0]']            32)                                                               conv2_1 (Conv3D)               (None, 64, 64, 32,   128032      ['down_conv1_prelu[0][0]']       32)                                                               conv2_1_prelu (PReLU)          (None, 64, 64, 32,   4194304     ['conv2_1[0][0]']                32)                                                               conv2_2 (Conv3D)               (None, 64, 64, 32,   128032      ['conv2_1_prelu[0][0]']          32)                                                               conv2_2_prelu (PReLU)          (None, 64, 64, 32,   4194304     ['conv2_2[0][0]']                32)                                                               concat2 (Concatenate)          (None, 64, 64, 32,   0           ['down_conv1_prelu[0][0]',       64)                               'conv2_2_prelu[0][0]']          down_conv2 (Conv3D)            (None, 32, 32, 16,   32832       ['concat2[0][0]']                64)                                                               down_conv2_prelu (PReLU)       (None, 32, 32, 16,   1048576     ['down_conv2[0][0]']             64)                                                               conv3_1 (Conv3D)               (None, 32, 32, 16,   512064      ['down_conv2_prelu[0][0]']       64)                                                               conv3_1_prelu (PReLU)          (None, 32, 32, 16,   1048576     ['conv3_1[0][0]']                64)                                                               conv3_2 (Conv3D)               (None, 32, 32, 16,   512064      ['conv3_1_prelu[0][0]']          64)                                                               conv3_2_prelu (PReLU)          (None, 32, 32, 16,   1048576     ['conv3_2[0][0]']                64)                                                               conv3_3 (Conv3D)               (None, 32, 32, 16,   512064      ['conv3_2_prelu[0][0]']          64)                                                               conv3_3_prelu (PReLU)          (None, 32, 32, 16,   1048576     ['conv3_3[0][0]']                64)                                                               concat3 (Concatenate)          (None, 32, 32, 16,   0           ['conv3_3_prelu[0][0]',          128)                              'down_conv2_prelu[0][0]']       down_conv3 (Conv3D)            (None, 16, 16, 8, 1  131200      ['concat3[0][0]']                28)                                                               down_conv3_prelu (PReLU)       (None, 16, 16, 8, 1  262144      ['down_conv3[0][0]']             28)                                                               conv4_1 (Conv3D)               (None, 16, 16, 8, 1  2048128     ['down_conv3_prelu[0][0]']       28)                                                               conv4_1_prelu (PReLU)          (None, 16, 16, 8, 1  262144      ['conv4_1[0][0]']                28)                                                               conv4_2 (Conv3D)               (None, 16, 16, 8, 1  2048128     ['conv4_1_prelu[0][0]']          28)                                                               conv4_2_prelu (PReLU)          (None, 16, 16, 8, 1  262144      ['conv4_2[0][0]']                28)                                                               conv4_3 (Conv3D)               (None, 16, 16, 8, 1  2048128     ['conv4_2_prelu[0][0]']          28)                                                               conv4_3_prelu (PReLU)          (None, 16, 16, 8, 1  262144      ['conv4_3[0][0]']                28)                                                               concat4 (Concatenate)          (None, 16, 16, 8, 2  0           ['down_conv3_prelu[0][0]',       56)                               'conv4_3_prelu[0][0]']          down_conv4 (Conv3D)            (None, 8, 8, 4, 256  524544      ['concat4[0][0]']                )                                                                 down_conv4_prelu (PReLU)       (None, 8, 8, 4, 256  65536       ['down_conv4[0][0]']             )                                                                 conv5_1 (Conv3D)               (None, 8, 8, 4, 256  8192256     ['down_conv4_prelu[0][0]']       )                                                                 conv5_1_prelu (PReLU)          (None, 8, 8, 4, 256  65536       ['conv5_1[0][0]']                )                                                                 conv5_2 (Conv3D)               (None, 8, 8, 4, 256  8192256     ['conv5_1_prelu[0][0]']          )                                                                 conv5_2_prelu (PReLU)          (None, 8, 8, 4, 256  65536       ['conv5_2[0][0]']                )                                                                 conv5_3 (Conv3D)               (None, 8, 8, 4, 256  8192256     ['conv5_2_prelu[0][0]']          )                                                                 conv5_3_prelu (PReLU)          (None, 8, 8, 4, 256  65536       ['conv5_3[0][0]']                )                                                                 concat5 (Concatenate)          (None, 8, 8, 4, 512  0           ['down_conv4_prelu[0][0]',       )                                 'conv5_3_prelu[0][0]']          up_conv5 (Conv3DTranspose)     (None, 16, 16, 8, 2  1048832     ['concat5[0][0]']                56)                                                               up_conv5_prelu (PReLU)         (None, 16, 16, 8, 2  524288      ['up_conv5[0][0]']               56)                                                               conv6 (Concatenate)            (None, 16, 16, 8, 5  0           ['up_conv5_prelu[0][0]',         12)                               'concat4[0][0]']                conv6_1 (Conv3D)               (None, 16, 16, 8, 2  16384256    ['conv6[0][0]']                  56)                                                               conv6_1_prelu (PReLU)          (None, 16, 16, 8, 2  524288      ['conv6_1[0][0]']                56)                                                               conv6_2 (Conv3D)               (None, 16, 16, 8, 2  8192256     ['conv6_1_prelu[0][0]']          56)                                                               conv6_2_prelu (PReLU)          (None, 16, 16, 8, 2  524288      ['conv6_2[0][0]']                56)                                                               conv6_3 (Conv3D)               (None, 16, 16, 8, 2  8192256     ['conv6_2_prelu[0][0]']          56)                                                               conv6_3_prelu (PReLU)          (None, 16, 16, 8, 2  524288      ['conv6_3[0][0]']                56)                                                               concat6 (Concatenate)          (None, 16, 16, 8, 5  0           ['up_conv5_prelu[0][0]',         12)                               'conv6_3_prelu[0][0]']          up_conv6 (Conv3DTranspose)     (None, 32, 32, 16,   524416      ['concat6[0][0]']                128)                                                              up_conv6_prelu (PReLU)         (None, 32, 32, 16,   2097152     ['up_conv6[0][0]']               128)                                                              conv7 (Concatenate)            (None, 32, 32, 16,   0           ['up_conv6_prelu[0][0]',         256)                              'concat3[0][0]']                conv7_1 (Conv3D)               (None, 32, 32, 16,   4096128     ['conv7[0][0]']                  128)                                                              conv7_1_prelu (PReLU)          (None, 32, 32, 16,   2097152     ['conv7_1[0][0]']                128)                                                              conv7_2 (Conv3D)               (None, 32, 32, 16,   2048128     ['conv7_1_prelu[0][0]']          128)                                                              conv7_2_prelu (PReLU)          (None, 32, 32, 16,   2097152     ['conv7_2[0][0]']                128)                                                              conv7_3 (Conv3D)               (None, 32, 32, 16,   2048128     ['conv7_2_prelu[0][0]']          128)                                                              conv7_3_prelu (PReLU)          (None, 32, 32, 16,   2097152     ['conv7_3[0][0]']                128)                                                              concat7 (Concatenate)          (None, 32, 32, 16,   0           ['up_conv6_prelu[0][0]',         256)                              'conv7_3_prelu[0][0]']          up_conv7 (Conv3DTranspose)     (None, 64, 64, 32,   131136      ['concat7[0][0]']                64)                                                               up_conv7_prelu (PReLU)         (None, 64, 64, 32,   8388608     ['up_conv7[0][0]']               64)                                                               conv8 (Concatenate)            (None, 64, 64, 32,   0           ['up_conv7_prelu[0][0]',         128)                              'concat2[0][0]']                conv8_1 (Conv3D)               (None, 64, 64, 32,   1024064     ['conv8[0][0]']                  64)                                                               conv8_1_prelu (PReLU)          (None, 64, 64, 32,   8388608     ['conv8_1[0][0]']                64)                                                               conv8_2 (Conv3D)               (None, 64, 64, 32,   512064      ['conv8_1_prelu[0][0]']          64)                                                               conv8_2_prelu (PReLU)          (None, 64, 64, 32,   8388608     ['conv8_2[0][0]']                64)                                                               concat8 (Concatenate)          (None, 64, 64, 32,   0           ['conv8_2_prelu[0][0]',          128)                              'up_conv7_prelu[0][0]']         up_conv8 (Conv3DTranspose)     (None, 128, 128, 64  32800       ['concat8[0][0]']                , 32)                                                             up_conv8_prelu (PReLU)         (None, 128, 128, 64  33554432    ['up_conv8[0][0]']               , 32)                                                             conv9 (Concatenate)            (None, 128, 128, 64  0           ['up_conv8_prelu[0][0]',         , 49)                             'concat1[0][0]']                conv9_1 (Conv3D)               (None, 128, 128, 64  196032      ['conv9[0][0]']                  , 32)                                                             concat9 (Concatenate)          (None, 128, 128, 64  0           ['conv9_1[0][0]',                , 64)                             'up_conv8_prelu[0][0]']         out (Conv3D)                   (None, 128, 128, 64  130         ['concat9[0][0]']                , 2)                                                              ==================================================================================================
Total params: 181,710,178
Trainable params: 181,710,178
Non-trainable params: 0
__________________________________________________________________________________________________Process finished with exit code 0