加载包含 Lambda 层的 Keras 模型时出现 AttributeError

Question

我正在尝试加载经过训练的 keras 模型 (SeResnext)，该模型架构也包含“Lambda”层。

现在当我尝试在我的脚本中加载模型时，我得到了这个属性错误：

Traceback (most recent call last):
  File "predict.py", line 9, in <module>
    model = keras.models.load_model('mySeResnextModel.hdf5')
  File "/usr/local/lib/python3.6/site-packages/keras/engine/saving.py", line 419, in load_model
    model = _deserialize_model(f, custom_objects, compile)
  File "/usr/local/lib/python3.6/site-packages/keras/engine/saving.py", line 225, in _deserialize_model
    model = model_from_config(model_config, custom_objects=custom_objects)
  File "/usr/local/lib/python3.6/site-packages/keras/engine/saving.py", line 458, in model_from_config
    return deserialize(config, custom_objects=custom_objects)
  File "/usr/local/lib/python3.6/site-packages/keras/layers/__init__.py", line 55, in deserialize
    printable_module_name='layer')
  File "/usr/local/lib/python3.6/site-packages/keras/utils/generic_utils.py", line 145, in deserialize_keras_object
    list(custom_objects.items())))
  File "/usr/local/lib/python3.6/site-packages/keras/engine/network.py", line 1022, in from_config
    process_layer(layer_data)
  File "/usr/local/lib/python3.6/site-packages/keras/engine/network.py", line 1008, in process_layer
    custom_objects=custom_objects)
  File "/usr/local/lib/python3.6/site-packages/keras/layers/__init__.py", line 55, in deserialize
    printable_module_name='layer')
  File "/usr/local/lib/python3.6/site-packages/keras/utils/generic_utils.py", line 145, in deserialize_keras_object
    list(custom_objects.items())))
  File "/usr/local/lib/python3.6/site-packages/keras/layers/core.py", line 732, in from_config
    printable_module_name='function in Lambda layer')
  File "/usr/local/lib/python3.6/site-packages/keras/utils/generic_utils.py", line 162, in deserialize_keras_object
    fn = module_objects.get(function_name)
AttributeError: 'NoneType' object has no attribute 'get'

我认为这是因为 Keras 没有任何名为 Lambda 的内置层，所以它根本无法识别该层。

现在当我搜索这个问题时，我得到的唯一但不是那么有益的解决方案是删除 lambda 层，但就我而言，它们太多了。我也可以导入然后将 Lambda 层用作自定义层吗？

我怎样才能找到解决这个问题的方法？

P.S.：我在搜索后找到了这个（https://github.com/keras-team/keras/issues/4871），如何在这里使用自定义对象参数？

Answer 1

实际上 Keras 有 Lambda 层（keras.layers.Lambda），但问题是由它使用的函数引起的。

要解决它，您可以使用 custom_objects 参数传递所需的函数，例如：

def channel_zeropad(x, channel_axis=3):
    '''
    Zero-padding for channle dimensions.
    Note that padded channles are added like (Batch, H, W, 2/x + x + 2/x).
    '''
    shape = list(x.shape)
    y = K.zeros_like(x)

    if channel_axis == 3:
        y = y[:, :, :, :shape[channel_axis] // 2]
    else:
        y = y[:, :shape[channel_axis] // 2, :, :]

    return concatenate([y, x, y], channel_axis)

def channel_zeropad_output(input_shape, channel_axis=3):
    '''
    Function for setting a channel dimension for zero padding.
    '''
    shape = list(input_shape)
    shape[channel_axis] *= 2

    return tuple(shape)

model = keras.models.load_model('mySeResnextModel.hdf5',
              custom_objects={'channel_zeropad': channel_zeropad,
                              'channel_zeropad_output': channel_zeropad_output})

或单独定义模型及其负载权重：

model = SEResNeXt().model  # if you are using senet-keras
model.load_weights('mySeResnextModel.hdf5', by_name=True)