如何将tensorflow张量转换为keras张量或调整keras特征图的大小？

Question

如何在keras中调整特征映射的大小或将tensorflow张量转换为keras张量？你知道吗

我想调整out of keras层的大小，并使用K.resize_images，但失败了。你知道吗

    block1_btchnorm2 = BatchNormalization(name ='b1_bn2')(block1_conv2)
    block1_conv3 = Conv2D(128, (3,3), activation='elu',name='b1_c3')(block1_btchnorm2)
    block1_btchnorm3 = BatchNormalization(name ='b1_bn3')(block1_conv3)
    block1_maxpooling =  MaxPooling2D(pool_size=(2,2),name ='b1_mp')(block1_btchnorm3)
    block1_out = K.resize_images(block1_maxpooling, height_factor =64/124 , width_factor = 64/124, data_format='channels_last')

AttributeError: 'Tensor' object has no attribute '_keras_history'

Answer 1

必须使用Lambda层将任何后端函数应用于keras张量：

block1_out = Lambda(lambda x: K.resize_images(x, height_factor =64/124 , width_factor = 64/124, data_format='channels_last'))(block1_maxpooling)