keras ValueError：生成器的输出应该是元组（x，y，sample_weight）或（x，y）。发现：无

Question

我有来自Retina Unet 的 Unet 模型，但是我已经增强了图像和蒙版。现在？它给了我这个错误ValueError: output of generator should be a tuple (x, y, sample_weight) or (x, y). Found: None我想训练增强（图像和掩码）并验证增强图像和掩码。

批量生成函数：

    def batch_generator(X_gen,Y_gen):
            yield(X_batch,Y_batch)



model = get_unet(1,img_width,img_hight)  #the U-net model
print("Model Summary")
print(model.summary())
print "Check: final output of the network:"
print model.output_shape

#============  Training ==================================
checkpointer = ModelCheckpoint(filepath='./'+'SAEED'+'_best_weights.h5', verbose=2, monitor='val_acc', mode='auto', save_best_only=True) #save at each epoch if the validation decreased
print("Now augumenting training")
datagen = ImageDataGenerator(rotation_range=120)
#traing augumentation.
train_images_generator = datagen.flow_from_directory(train_images_dir,target_size=(img_width,img_hight),batch_size=30,class_mode=None)
train_mask_generator = datagen.flow_from_directory(train_masks_dir,target_size=(img_width,img_hight),batch_size=30,class_mode=None)
print("Now augumenting val")
#val augumentation.
val_images_generator = datagen.flow_from_directory(val_images_dir,target_size=(img_width,img_hight),batch_size=30,class_mode=None)
val_masks_generator = datagen.flow_from_directory(val_masks_dir,target_size=(img_width,img_hight),batch_size=30,class_mode=None)

print("Now augumenting test")
#test augumentation
test_images_generator = datagen.flow_from_directory(test_images_dir,target_size=(img_width,img_hight),batch_size=25,class_mode=None)
test_masks_generator = datagen.flow_from_directory(test_masks_dir,target_size=(img_width,img_hight),batch_size=25,class_mode=None)
#fitting model.
print("Now fitting the model ")
#model.fit_generator(train_generator,samples_per_epoch = nb_train_samples*2,nb_epoch=nb_epoch,validation_data=val_generator,nb_val_samples=nb_val_samples,callbacks=[checkpointer])
print("train_images_generator size {} and type is {}".format(next(train_images_generator).shape,type(next(train_images_generator))))
print("train_masks_generator size {} and type is {}".format(next(train_mask_generator).shape,type(next(train_mask_generator))))

model.fit_generator(batch_generator(train_images_generator,train_mask_generator),samples_per_epoch = nb_train_samples,nb_epoch=nb_epoch,validation_data=batch_generator(val_images_generator,val_masks_generator),nb_val_samples=nb_val_samples,callbacks=[checkpointer])
print("Finished fitting the model")

` 模型总结：

`

Model Summary
____________________________________________________________________________________________________
Layer (type)                     Output Shape          Param #     Connected to
====================================================================================================
input_1 (InputLayer)             (None, 1, 160, 160)   0
____________________________________________________________________________________________________
convolution2d_1 (Convolution2D)  (None, 32, 160, 160)  320         input_1[0][0]
____________________________________________________________________________________________________
dropout_1 (Dropout)              (None, 32, 160, 160)  0           convolution2d_1[0][0]
____________________________________________________________________________________________________
convolution2d_2 (Convolution2D)  (None, 32, 160, 160)  9248        dropout_1[0][0]
____________________________________________________________________________________________________
maxpooling2d_1 (MaxPooling2D)    (None, 32, 80, 80)    0           convolution2d_2[0][0]
____________________________________________________________________________________________________
convolution2d_3 (Convolution2D)  (None, 64, 80, 80)    18496       maxpooling2d_1[0][0]
____________________________________________________________________________________________________
dropout_2 (Dropout)              (None, 64, 80, 80)    0           convolution2d_3[0][0]
____________________________________________________________________________________________________
convolution2d_4 (Convolution2D)  (None, 64, 80, 80)    36928       dropout_2[0][0]
____________________________________________________________________________________________________
maxpooling2d_2 (MaxPooling2D)    (None, 64, 40, 40)    0           convolution2d_4[0][0]
____________________________________________________________________________________________________
convolution2d_5 (Convolution2D)  (None, 128, 40, 40)   73856       maxpooling2d_2[0][0]
____________________________________________________________________________________________________
dropout_3 (Dropout)              (None, 128, 40, 40)   0           convolution2d_5[0][0]
____________________________________________________________________________________________________
convolution2d_6 (Convolution2D)  (None, 128, 40, 40)   147584      dropout_3[0][0]
____________________________________________________________________________________________________
upsampling2d_1 (UpSampling2D)    (None, 128, 80, 80)   0           convolution2d_6[0][0]
____________________________________________________________________________________________________
merge_1 (Merge)                  (None, 192, 80, 80)   0           upsampling2d_1[0][0]
                                                                   convolution2d_4[0][0]
____________________________________________________________________________________________________
convolution2d_7 (Convolution2D)  (None, 64, 80, 80)    110656      merge_1[0][0]
____________________________________________________________________________________________________
dropout_4 (Dropout)              (None, 64, 80, 80)    0           convolution2d_7[0][0]
____________________________________________________________________________________________________
convolution2d_8 (Convolution2D)  (None, 64, 80, 80)    36928       dropout_4[0][0]
____________________________________________________________________________________________________
upsampling2d_2 (UpSampling2D)    (None, 64, 160, 160)  0           convolution2d_8[0][0]
____________________________________________________________________________________________________
merge_2 (Merge)                  (None, 96, 160, 160)  0           upsampling2d_2[0][0]
                                                                   convolution2d_2[0][0]
____________________________________________________________________________________________________
convolution2d_9 (Convolution2D)  (None, 32, 160, 160)  27680       merge_2[0][0]
____________________________________________________________________________________________________
dropout_5 (Dropout)              (None, 32, 160, 160)  0           convolution2d_9[0][0]
____________________________________________________________________________________________________
convolution2d_10 (Convolution2D) (None, 32, 160, 160)  9248        dropout_5[0][0]
____________________________________________________________________________________________________
convolution2d_11 (Convolution2D) (None, 2, 160, 160)   66          convolution2d_10[0][0]
____________________________________________________________________________________________________
reshape_1 (Reshape)              (None, 2, 25600)      0           convolution2d_11[0][0]
____________________________________________________________________________________________________
permute_1 (Permute)              (None, 25600, 2)      0           reshape_1[0][0]
____________________________________________________________________________________________________
activation_1 (Activation)        (None, 25600, 2)      0           permute_1[0][0]
====================================================================================================
Total params: 471,010
Trainable params: 471,010
Non-trainable params: 0

`

有什么想法吗？谢谢。

Answer 1

以防以后有人遇到同样的问题。

问题是生成器问题。固定在下面

def batch_generator(X_gen,Y_gen): while true: yield(X_gen.next(),Y_gen.next())

Answer 2

在我的例子中，将 class_mode 添加到生成器解决了这个问题。 例如：

train_generator = train_datagen.flow_from_directory(
    train_dir,
    target_size=(image_size, image_size),
    batch_size=batch_size,
    class_mode='categorical')

您可以选择：

• 二进制：二进制标签的一维 numpy 数组

• categorical : one-hot 编码标签的二维 numpy 数组。支持多标签输出。

• sparse : 1D numpy 整数标签数组

• 输入：与输入图像相同的图像（主要用于自动编码器）

• 其他：y_col 数据的 numpy 数组

Btw None 应该也可以工作..但这对我来说是解决方案