CNN图像分类训练报错‘boundary mode not supported’求助

问题背景 #

跟着教程做图像分类，用CNN构建了如下模型：

model=tf.keras.models.Sequential([
    #first layer an input layer a shape of 100*150 RGP array of pictures 
    # 2D convutional layer with 32 nodes 3*3 filter
    tf.keras.layers.Conv2D(32,(3,3),activation='relu',input_shape=(100,150,3)),
    #2d maxpooling with size of 2*2
    tf.keras.layers.MaxPooling2D(2,2),
    
    #second layer
    tf.keras.layers.Conv2D(64,(3,3),activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),
    
    #third layer 
    tf.keras.layers.Conv2D(128,(3,3),activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),
    
    #forth layer 
    tf.keras.layers.Conv2D(256,(3,3),activation='relu'),
    tf.keras.layers.MaxPooling2D(2,2),
    
    #flatten layer 
    tf.keras.layers.Flatten(),
    
    #dense layer 
    tf.keras.layers.Dense(512,activation='relu'),
    tf.keras.layers.Dense(3,activation='softmax')
])
model.compile(loss='categorical_crossentropy',optimizer=tf.optimizers.Adam(),
              metrics=['accuracy'])

运行训练代码时触发错误：

history=model.fit(train_gen,steps_per_epoch=25,epochs=20,
                  validation_data=validation_gen,validation_steps=5,verbose=2,
                  callbacks=[my_callback()]
)

报错回溯：

---------------------------------------------------------------------------
RuntimeError                              Traceback (most recent call last)
c:\Users\rabee\OneDrive\Desktop\rps-final-dataset\rbs_classification.ipynb Cell 20 in <cell line: 1>()
----> 1 model.fit(train_gen,steps_per_epoch=25,epochs=20,
      2                   validation_data=validation_gen,validation_steps=5,verbose=2,
      3                   callbacks=[my_callback()]
      4     
      5 )

File c:\Users\rabee\AppData\Local\Programs\Python\Python310\lib\site-packages\keras\utils\traceback_utils.py:67, in filter_traceback.<locals>.error_handler(*args, **kwargs)
     65 except Exception as e:  # pylint: disable=broad-except
     66   filtered_tb = _process_traceback_frames(e.__traceback__)
---> 67   raise e.with_traceback(filtered_tb) from None
     68 finally:
     69   del filtered_tb

File c:\Users\rabee\AppData\Local\Programs\Python\Python310\lib\site-packages\scipy\ndimage\_interpolation.py:574, in affine_transform(input, matrix, offset, output_shape, output, order, mode, cval, prefilter)
    572     npad = 0
    573     filtered = input
--> 574 mode = _ni_support._extend_mode_to_code(mode)
    575 matrix = numpy.asarray(matrix, dtype=numpy.float64)
    576 if matrix.ndim not in [1, 2] or matrix.shape[0] < 1:

File c:\Users\rabee\AppData\Local\Programs\Python\Python310\lib\site-packages\scipy\ndimage\_ni_support.py:54, in _extend_mode_to_code(mode)
     52     return 6
     53 else:
---> 54     raise RuntimeError('boundary mode not supported')

RuntimeError: boundary mode not supported

错误原因 #

这个错误根源在图像生成器的预处理参数，具体是scipy的ndimage模块不识别你设置的边界填充模式。比如用ImageDataGenerator时fill_mode设了scipy不支持的值，或者自定义生成器里调用scipy图像函数时传了无效的mode参数。

解决方法 #

• 检查ImageDataGenerator的fill_mode参数
  确保fill_mode用scipy支持的取值：'nearest'、'reflect'、'constant'、'wrap'。比如把生成器改成这样：

  from tensorflow.keras.preprocessing.image import ImageDataGenerator
  
  # 训练集生成器
  train_gen = ImageDataGenerator(
      rescale=1./255,
      fill_mode='nearest'  # 替换成支持的模式，别用'mirror'这类scipy不认的
  ).flow_from_directory(
      # 你的路径和其他参数
  )
  
  # 验证集同理
  validation_gen = ImageDataGenerator(
      rescale=1./255,
      fill_mode='nearest'
  ).flow_from_directory(
      # 你的路径和其他参数
  )
  

• 排查自定义数据生成器
  如果是自己写的生成器，在做图像变换（缩放、旋转等）调用scipy函数时，确保mode参数是上述支持的值，别传拼写错误或不兼容的字符串。
• 升级scipy版本
  版本不兼容也可能导致这个问题，执行命令升级到最新稳定版：

  pip install --upgrade scipy
  

• 检查自定义回调函数
  你的my_callback()如果涉及图像处理逻辑，也要排查里面有没有用到不支持的边界模式参数。

内容的提问来源于stack exchange，提问作者noob