解决keras model 多模型使用的方法

Tensor is not an element of this graph; deploying Keras model

解决办法：
在每个模型实例化时创建新的graph
创建模型之前 添加graph
部分代码

class Inference(object):
    loaded_model = None
    classes_in_keras_format = None
    graph=None
    session=None

    def __init__(self, company_id):
        self.company_id = company_id
        if Inference.classes_in_keras_format is not None:
            return
        self.load_classes()
        Inference.classes_in_keras_format = dict(zip(config.classes, range(len(config.classes))))
        self.set_img_format()
        Inference.session = tf.Session()
        Inference.graph = tf.get_default_graph()
        with Inference.graph.as_default():
            with Inference.session.as_default():
                self.create_model()
                self.load_model()
        
	def predict(self, predict_images):
        with Inference.graph.as_default():
            with Inference.session.as_default():
                predict_images = self.preprocess_input(predict_images)
                out = Inference.loaded_model.predict(np.array(predict_images))

解决思路

解决多模型方法 graph
https://stackoverflow.com/questions/51127344/tensor-is-not-an-element-of-this-graph-deploying-keras-model?r=SearchResults