Я пытался изменить se gnet с объединением индексов в keras, но когда я в четыре раза увеличил количество ядер conv для каждого блока conv, во время обучения все пошло не так. Не знаю почему. Не могли бы вы мне помочь? Заранее спасибо.
трассировка
тензорный поток. python .framework.errors_impl.InvalidArgumentError: Несовместимые формы: [2,64,64,256] vs. [512] [[{ {node replica_0 / model_3 / max_unpooling2d_2 / max_unpooling2d_2 / mul_3}}]] [[{{node loss / mul}}]]
model.summary ()
Layer (type) Output Shape Param # Connected to
==================================================================================================
input_1 (InputLayer) (None, 512, 512, 3) 0
__________________________________________________________________________________________________
conv2d_1 (Conv2D) (None, 512, 512, 64) 1792 input_1[0][0]
__________________________________________________________________________________________________
batch_normalization_1 (BatchNor (None, 512, 512, 64) 256 conv2d_1[0][0]
__________________________________________________________________________________________________
conv2d_2 (Conv2D) (None, 512, 512, 64) 36928 batch_normalization_1[0][0]
__________________________________________________________________________________________________
batch_normalization_2 (BatchNor (None, 512, 512, 64) 256 conv2d_2[0][0]
__________________________________________________________________________________________________
max_pooling_with_argmax2d_1 (Ma [(None, 256, 256, 64 0 batch_normalization_2[0][0]
__________________________________________________________________________________________________
conv2d_3 (Conv2D) (None, 256, 256, 128 73856 max_pooling_with_argmax2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_3 (BatchNor (None, 256, 256, 128 512 conv2d_3[0][0]
__________________________________________________________________________________________________
conv2d_4 (Conv2D) (None, 256, 256, 128 147584 batch_normalization_3[0][0]
__________________________________________________________________________________________________
batch_normalization_4 (BatchNor (None, 256, 256, 128 512 conv2d_4[0][0]
__________________________________________________________________________________________________
max_pooling_with_argmax2d_2 (Ma [(None, 128, 128, 12 0 batch_normalization_4[0][0]
__________________________________________________________________________________________________
conv2d_5 (Conv2D) (None, 128, 128, 256 295168 max_pooling_with_argmax2d_2[0][0]
__________________________________________________________________________________________________
batch_normalization_5 (BatchNor (None, 128, 128, 256 1024 conv2d_5[0][0]
__________________________________________________________________________________________________
conv2d_6 (Conv2D) (None, 128, 128, 256 590080 batch_normalization_5[0][0]
__________________________________________________________________________________________________
batch_normalization_6 (BatchNor (None, 128, 128, 256 1024 conv2d_6[0][0]
__________________________________________________________________________________________________
max_pooling_with_argmax2d_3 (Ma [(None, 64, 64, 256) 0 batch_normalization_6[0][0]
__________________________________________________________________________________________________
conv2d_7 (Conv2D) (None, 64, 64, 512) 1180160 max_pooling_with_argmax2d_3[0][0]
__________________________________________________________________________________________________
batch_normalization_7 (BatchNor (None, 64, 64, 512) 2048 conv2d_7[0][0]
__________________________________________________________________________________________________
conv2d_8 (Conv2D) (None, 64, 64, 512) 2359808 batch_normalization_7[0][0]
__________________________________________________________________________________________________
batch_normalization_8 (BatchNor (None, 64, 64, 512) 2048 conv2d_8[0][0]
__________________________________________________________________________________________________
max_pooling_with_argmax2d_4 (Ma [(None, 32, 32, 512) 0 batch_normalization_8[0][0]
__________________________________________________________________________________________________
max_unpooling2d_1 (MaxUnpooling (None, 64, 64, 512) 0 max_pooling_with_argmax2d_4[0][0]
max_pooling_with_argmax2d_4[0][1]
__________________________________________________________________________________________________
conv2d_13 (Conv2D) (None, 64, 64, 512) 2359808 max_unpooling2d_1[0][0]
__________________________________________________________________________________________________
batch_normalization_9 (BatchNor (None, 64, 64, 512) 2048 conv2d_13[0][0]
__________________________________________________________________________________________________
conv2d_14 (Conv2D) (None, 64, 64, 512) 2359808 batch_normalization_9[0][0]
__________________________________________________________________________________________________
batch_normalization_10 (BatchNo (None, 64, 64, 512) 2048 conv2d_14[0][0]
__________________________________________________________________________________________________
conv2d_15 (Conv2D) (None, 64, 64, 512) 2359808 batch_normalization_10[0][0]
__________________________________________________________________________________________________
batch_normalization_11 (BatchNo (None, 64, 64, 512) 2048 conv2d_15[0][0]
__________________________________________________________________________________________________
max_unpooling2d_2 (MaxUnpooling (None, 128, 128, 256 0 batch_normalization_11[0][0]
max_pooling_with_argmax2d_3[0][1]
__________________________________________________________________________________________________
conv2d_16 (Conv2D) (None, 128, 128, 256 590080 max_unpooling2d_2[0][0]
__________________________________________________________________________________________________
batch_normalization_12 (BatchNo (None, 128, 128, 256 1024 conv2d_16[0][0]
__________________________________________________________________________________________________
conv2d_17 (Conv2D) (None, 128, 128, 256 590080 batch_normalization_12[0][0]
__________________________________________________________________________________________________
batch_normalization_13 (BatchNo (None, 128, 128, 256 1024 conv2d_17[0][0]
__________________________________________________________________________________________________
max_unpooling2d_3 (MaxUnpooling (None, 256, 256, 128 0 batch_normalization_13[0][0]
max_pooling_with_argmax2d_2[0][1]
__________________________________________________________________________________________________
conv2d_18 (Conv2D) (None, 256, 256, 128 147584 max_unpooling2d_3[0][0]
__________________________________________________________________________________________________
batch_normalization_14 (BatchNo (None, 256, 256, 128 512 conv2d_18[0][0]
__________________________________________________________________________________________________
conv2d_19 (Conv2D) (None, 256, 256, 128 147584 batch_normalization_14[0][0]
__________________________________________________________________________________________________
batch_normalization_15 (BatchNo (None, 256, 256, 128 512 conv2d_19[0][0]
__________________________________________________________________________________________________
max_unpooling2d_4 (MaxUnpooling (None, 512, 512, 64) 0 batch_normalization_15[0][0]
max_pooling_with_argmax2d_1[0][1]
__________________________________________________________________________________________________
conv2d_20 (Conv2D) (None, 512, 512, 64) 36928 max_unpooling2d_4[0][0]
__________________________________________________________________________________________________
batch_normalization_16 (BatchNo (None, 512, 512, 64) 256 conv2d_20[0][0]
__________________________________________________________________________________________________
conv2d_9 (Conv2D) (None, 512, 512, 1) 65 batch_normalization_2[0][0]
__________________________________________________________________________________________________
conv2d_10 (Conv2D) (None, 256, 256, 1) 129 batch_normalization_4[0][0]
__________________________________________________________________________________________________
conv2d_11 (Conv2D) (None, 128, 128, 1) 257 batch_normalization_6[0][0]
__________________________________________________________________________________________________
conv2d_12 (Conv2D) (None, 64, 64, 1) 513 batch_normalization_8[0][0]
__________________________________________________________________________________________________
conv2d_21 (Conv2D) (None, 64, 64, 1) 513 batch_normalization_10[0][0]
__________________________________________________________________________________________________
conv2d_22 (Conv2D) (None, 128, 128, 1) 257 batch_normalization_12[0][0]
__________________________________________________________________________________________________
conv2d_23 (Conv2D) (None, 256, 256, 1) 129 batch_normalization_14[0][0]
__________________________________________________________________________________________________
conv2d_24 (Conv2D) (None, 512, 512, 1) 65 batch_normalization_16[0][0]
__________________________________________________________________________________________________
conv2d_transpose_1 (Conv2DTrans (None, 512, 512, 1) 2 conv2d_9[0][0]
__________________________________________________________________________________________________
conv2d_transpose_2 (Conv2DTrans (None, 512, 512, 1) 5 conv2d_10[0][0]
__________________________________________________________________________________________________
conv2d_transpose_3 (Conv2DTrans (None, 512, 512, 1) 17 conv2d_11[0][0]
__________________________________________________________________________________________________
conv2d_transpose_4 (Conv2DTrans (None, 512, 512, 1) 65 conv2d_12[0][0]
__________________________________________________________________________________________________
conv2d_transpose_5 (Conv2DTrans (None, 512, 512, 1) 65 conv2d_21[0][0]
__________________________________________________________________________________________________
conv2d_transpose_6 (Conv2DTrans (None, 512, 512, 1) 17 conv2d_22[0][0]
__________________________________________________________________________________________________
conv2d_transpose_7 (Conv2DTrans (None, 512, 512, 1) 5 conv2d_23[0][0]
__________________________________________________________________________________________________
conv2d_transpose_8 (Conv2DTrans (None, 512, 512, 1) 2 conv2d_24[0][0]
__________________________________________________________________________________________________
concatenate_1 (Concatenate) (None, 512, 512, 4) 0 conv2d_transpose_1[0][0]
conv2d_transpose_2[0][0]
conv2d_transpose_3[0][0]
conv2d_transpose_4[0][0]
__________________________________________________________________________________________________
concatenate_2 (Concatenate) (None, 512, 512, 4) 0 conv2d_transpose_5[0][0]
conv2d_transpose_6[0][0]
conv2d_transpose_7[0][0]
conv2d_transpose_8[0][0]
__________________________________________________________________________________________________
concatenate_3 (Concatenate) (None, 512, 512, 8) 0 concatenate_1[0][0]
concatenate_2[0][0]
__________________________________________________________________________________________________
conv2d_25 (Conv2D) (None, 512, 512, 2) 18 concatenate_3[0][0]
__________________________________________________________________________________________________
reshape_1 (Reshape) (None, 262144, 2) 0 conv2d_25[0][0]
__________________________________________________________________________________________________
activation_1 (Activation) (None, 262144, 2) 0 reshape_1[0][0]
==================================================================================================
Total params: 13,296,332
Trainable params: 13,287,756
Non-trainable params: 8,576
keras 2.2.4 тензор потока 1.13.1