keras load_model在加载包含自定义层时会出现unknown layer的错误

解决方法：

1.在自定义层的定义文件YourLayer.py中添加如下代码：

def create_custom_objects():

    instance_holder = {"instance": None}

    class ClassWrapper(YourLayerName):

        def __init__(self, *args, **kwargs):

            instance_holder["instance"] = self

            super(ClassWrapper, self).__init__(*args, **kwargs)

    return {"ClassWrapper": ClassWrapper ,"YourLayerName": ClassWrapper}

2.然后加载模型

from YourLayer import create_custom_objects

model = keras.models.load_model(model_path, custom_objects=create_custom_objects())

参阅链接

python - Not able to load keras trained model - Stack Overflow