不兼容的形状：[32,1,100,100]与[32,1,96,96]

Question

嗨，我在代码中遇到了一个错误：不兼容的形状：[32,1,100,100]与[32,1,96,96]

CAE类（对象）： def init （自我，img_shape，actf ='relu'， learning_rate = 0.001，drop_rate = 0.5，do_batch_norm = True，do_drop = True）： 是否参加退学 '''

    self.learning_rate = learning_rate
    self.actf = actf
    self.img_shape = img_shape
    self.drop_rate = drop_rate
    self.do_batch_norm = do_batch_norm
    self.do_drop = do_drop

    self.model = self.build_model()

# just enter input function
def cae_data_input(self, x):
    return x

# input shape
def cae_shape(self, input_shape):
    return (input_shape[0], input_shape[1], input_shape[2])


# encoder part
def encoding_path(self, model):
    # encoder layer1
    model.add(Conv2D(16, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(MaxPooling2D((2,2), strides = (2,2)))

    # encoder layer2
    model.add(Conv2D(32, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(MaxPooling2D((2,2), strides = (2,2)))

    # encoder layer 3
    model.add(Conv2D(64, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(MaxPooling2D((2,2), strides = (2,2)))

    return model

# decoder part
def decoding_path(self, model):
    # decoder layer1
    model.add(Conv2D(64, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(UpSampling2D((2,2)))

    # decoder layer2
    model.add(Conv2D(32, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(UpSampling2D((2,2)))

    # decoder layer3
    model.add(Conv2D(16, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(BatchNormalization()) if self.do_batch_norm else None
    model.add(Activation(self.actf))
    model.add(UpSampling2D((2,2)))

    model.add(Conv2D(1, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(Activation('sigmoid'))

    return model

# build network
def build_model(self):
    model = models.Sequential()

    #  input layer
    model.add(Lambda(self.cae_data_input, input_shape = self.img_shape))

    # encoding
    model = self.encoding_path(model)

    # middle layer
    model.add(Conv2D(128, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(Dropout(self.drop_rate)) if self.do_drop else None

    # decoding
    model = self.decoding_path(model)

    model.compile(loss = 'binary_crossentropy', optimizer = Adam(lr = self.learning_rate)
                      , metrics = [dice_coef])
    return model

# train model
# def train(self, data, epoch = 10, batch_size = 32, val_split = 0.2, shuffle = True):

#     self.history = self.model.fit_generator(data,epochs = epoch)
#     return self.history
 
def train(self, X_train, Y_train, epoch = 10, batch_size = 32, val_split = 0.2, shuffle = True):

    self.history = self.model.fit(X_train, Y_train, validation_split = val_split,
                                      epochs = epoch, batch_size = batch_size, shuffle =  shuffle)
    return self.history






def build_model(self):
    model = models.Sequential()

    #  input layer
    model.add(Lambda(self.cae_data_input, input_shape = self.img_shape))

    # encoding
    model = self.encoding_path(model)

    # middle layer
    model.add(Conv2D(128, (3,3), strides = 1,  padding = 'same', kernel_initializer = 'he_normal'))
    model.add(Dropout(self.drop_rate)) if self.do_drop else None

     # decoding
    model = self.decoding_path(model)
    model.compile(loss ="binary_crossentropy", optimizer = Adam(lr = self.learning_rate)
                      , metrics = [dice_coef])
    return model

# train model
def train(self, x,y, epoch = 100, batch_size = 32, val_split = 0.2):

    self.history = self.model.fit(x,y,callbacks=callbacks)
    return self.history

火车和测试的大小：

print(trainX.shape, trainY.shape)
print(testX.shape, testY.shape)

产生

(2480, 100, 100, 1) (2480, 100, 100, 1)
(621, 100, 100, 1) (621, 100, 100, 1)