CRFNet(CameraRadarFusionNet)代码测试

经过对原作者的代码修改，可以用到3080平台。
请在git clone后将root文件夹名称修改为crfnet。

• 环境:

  • GPU: NVDIA Geforce RTX3080 x 2
  • System: Windows10
  • python: 3.6.13
  • tensorflow: 2.4.1
  • keras: 2.2.4

https://github.com/nacayu/CRFNet_reproduce_on_tensorflow2.4.4_3080
感谢作者提供了一个非常棒的融合模型！

修改不易，欢迎star原作者的工作（和我的修改代码）

---

模型结构：

__________________________________________________________________________________________________
Layer (type)                    Output Shape         Param #     Connected to
==================================================================================================
input_1 (InputLayer)            (None, 360, 640, 5)  0
__________________________________________________________________________________________________
image_channels (Lambda)         (None, 360, 640, 3)  0           input_1[0][0]
__________________________________________________________________________________________________
radar_channels (Lambda)         (None, 360, 640, 2)  0           input_1[0][0]
__________________________________________________________________________________________________
concat_0 (Concatenate)          (None, 360, 640, 5)  0           image_channels[0][0]
                                                                 radar_channels[0][0]
__________________________________________________________________________________________________
block1_conv1 (Conv2D)           (None, 360, 640, 64) 2944        concat_0[0][0]
__________________________________________________________________________________________________
block1_conv2 (Conv2D)           (None, 360, 640, 64) 36928       block1_conv1[0][0]
__________________________________________________________________________________________________
block1_pool (MaxPooling2D)      (None, 180, 320, 64) 0           block1_conv2[0][0]
__________________________________________________________________________________________________
rad_block1_pool (MaxPooling2D)  (None, 180, 320, 2)  0           radar_channels[0][0]
__________________________________________________________________________________________________
concat_1 (Concatenate)          (None, 180, 320, 66) 0           block1_pool[0][0]
                                                                 rad_block1_pool[0][0]
__________________________________________________________________________________________________
block2_conv1 (Conv2D)           (None, 180, 320, 128 76160       concat_1[0][0]
__________________________________________________________________________________________________
block2_conv2 (Conv2D)           (None, 180, 320, 128 147584      block2_conv1[0][0]
__________________________________________________________________________________________________
block2_pool (MaxPooling2D)      (None, 90, 160, 128) 0           block2_conv2[0][0]
__________________________________________________________________________________________________
rad_block2_pool (MaxPooling2D)  (None, 90, 160, 2)   0           rad_block1_pool[0][0]
__________________________________________________________________________________________________
concat_2 (Concatenate)          (None, 90, 160, 130) 0           block2_pool[0][0]
                                                                 rad_block2_pool[0][0]
__________________________________________________________________________________________________
block3_conv1 (Conv2D)           (None, 90, 160, 256) 299776      concat_2[0][0]
__________________________________________________________________________________________________
block3_conv2 (Conv2D)           (None, 90, 160, 256) 590080      block3_conv1[0][0]
__________________________________________________________________________________________________
block3_conv3 (Conv2D)           (None, 90, 160, 256) 590080      block3_conv2[0][0]
__________________________________________________________________________________________________
block3_pool (MaxPooling2D)      (None, 45, 80, 256)  0           block3_conv3[0][0]
__________________________________________________________________________________________________
rad_block3_pool (MaxPooling2D)  (None, 45, 80, 2)    0           rad_block2_pool[0][0]
__________________________________________________________________________________________________
concat_3 (Concatenate)          (None, 45, 80, 258)  0           block3_pool[0][0]
                                                                 rad_block3_pool[0][0]
__________________________________________________________________________________________________
block4_conv1 (Conv2D)           (None, 45, 80, 512)  1189376     concat_3[0][0]
__________________________________________________________________________________________________
block4_conv2 (Conv2D)           (None, 45, 80, 512)  2359808     block4_conv1[0][0]
__________________________________________________________________________________________________
block4_conv3 (Conv2D)           (None, 45, 80, 512)  2359808     block4_conv2[0][0]
__________________________________________________________________________________________________
block4_pool (MaxPooling2D)      (None, 22, 40, 512)  0           block4_conv3[0][0]
__________________________________________________________________________________________________
rad_block4_pool (MaxPooling2D)  (None, 22, 40, 2)    0           rad_block3_pool[0][0]
__________________________________________________________________________________________________
concat_4 (Concatenate)          (None, 22, 40, 514)  0           block4_pool[0][0]
                                                                 rad_block4_pool[0][0]
__________________________________________________________________________________________________
block5_conv1 (Conv2D)           (None, 22, 40, 512)  2369024     concat_4[0][0]
__________________________________________________________________________________________________
block5_conv2 (Conv2D)           (None, 22, 40, 512)  2359808     block5_conv1[0][0]
__________________________________________________________________________________________________
block5_conv3 (Conv2D)           (None, 22, 40, 512)  2359808     block5_conv2[0][0]
__________________________________________________________________________________________________
block5_pool (MaxPooling2D)      (None, 11, 20, 512)  0           block5_conv3[0][0]
__________________________________________________________________________________________________
rad_block5_pool (MaxPooling2D)  (None, 11, 20, 2)    0           rad_block4_pool[0][0]
__________________________________________________________________________________________________
concat_5 (Concatenate)          (None, 11, 20, 514)  0           block5_pool[0][0]
                                                                 rad_block5_pool[0][0]
__________________________________________________________________________________________________
C5_reduced (Conv2D)             (None, 11, 20, 254)  130810      concat_5[0][0]
__________________________________________________________________________________________________
P5_upsampled (UpsampleLike)     (None, 22, 40, 254)  0           C5_reduced[0][0]
                                                                 concat_4[0][0]
__________________________________________________________________________________________________
C4_reduced (Conv2D)             (None, 22, 40, 254)  130810      concat_4[0][0]
__________________________________________________________________________________________________
P4_merged (Add)                 (None, 22, 40, 254)  0           P5_upsampled[0][0]
                                                                 C4_reduced[0][0]
__________________________________________________________________________________________________
P4_upsampled (UpsampleLike)     (None, 45, 80, 254)  0           P4_merged[0][0]
                                                                 concat_3[0][0]
__________________________________________________________________________________________________
C3_reduced (Conv2D)             (None, 45, 80, 254)  65786       concat_3[0][0]
__________________________________________________________________________________________________
P6 (Conv2D)                     (None, 6, 10, 254)   1175258     concat_5[0][0]
__________________________________________________________________________________________________
P3_merged (Add)                 (None, 45, 80, 254)  0           P4_upsampled[0][0]
                                                                 C3_reduced[0][0]
__________________________________________________________________________________________________
rad_block6_pool (MaxPooling2D)  (None, 6, 10, 2)     0           rad_block5_pool[0][0]
__________________________________________________________________________________________________
C6_relu (Activation)            (None, 6, 10, 254)   0           P6[0][0]
__________________________________________________________________________________________________
P3 (Conv2D)                     (None, 45, 80, 254)  580898      P3_merged[0][0]
__________________________________________________________________________________________________
P4 (Conv2D)                     (None, 22, 40, 254)  580898      P4_merged[0][0]
__________________________________________________________________________________________________
P5 (Conv2D)                     (None, 11, 20, 254)  580898      C5_reduced[0][0]
__________________________________________________________________________________________________
P7 (Conv2D)                     (None, 3, 5, 254)    580898      C6_relu[0][0]
__________________________________________________________________________________________________
rad_block7_pool (MaxPooling2D)  (None, 3, 5, 2)      0           rad_block6_pool[0][0]
__________________________________________________________________________________________________
P3_rad (Concatenate)            (None, 45, 80, 256)  0           P3[0][0]
                                                                 rad_block3_pool[0][0]
__________________________________________________________________________________________________
P4_rad (Concatenate)            (None, 22, 40, 256)  0           P4[0][0]
                                                                 rad_block4_pool[0][0]
__________________________________________________________________________________________________
P5_rad (Concatenate)            (None, 11, 20, 256)  0           P5[0][0]
                                                                 rad_block5_pool[0][0]
__________________________________________________________________________________________________
P6_rad (Concatenate)            (None, 6, 10, 256)   0           P6[0][0]
                                                                 rad_block6_pool[0][0]
__________________________________________________________________________________________________
P7_rad (Concatenate)            (None, 3, 5, 256)    0           P7[0][0]
                                                                 rad_block7_pool[0][0]
__________________________________________________________________________________________________
regression_submodel (Model)     multiple             2443300     P3_rad[0][0]
                                                                 P4_rad[0][0]
                                                                 P5_rad[0][0]
                                                                 P6_rad[0][0]
                                                                 P7_rad[0][0]
__________________________________________________________________________________________________
classification_submodel (Model) multiple             2526280     P3_rad[0][0]
                                                                 P4_rad[0][0]
                                                                 P5_rad[0][0]
                                                                 P6_rad[0][0]
                                                                 P7_rad[0][0]
__________________________________________________________________________________________________
regression (Concatenate)        (None, 42975, 4)     0           regression_submodel[1][0]
                                                                 regression_submodel[2][0]
                                                                 regression_submodel[3][0]
                                                                 regression_submodel[4][0]
                                                                 regression_submodel[5][0]
__________________________________________________________________________________________________
classification (Concatenate)    (None, 42975, 8)     0           classification_submodel[1][0]
                                                                 classification_submodel[2][0]
                                                                 classification_submodel[3][0]
                                                                 classification_submodel[4][0]
                                                                 classification_submodel[5][0]
==================================================================================================
Total params: 23,537,020
Trainable params: 23,537,020
Non-trainable params: 0
__________________________________________________________________________________________________