使用Tensorflow检查点在C ++中还原模型 [英] Using Tensorflow checkpoint to restore model in C++

问题描述

我已经训练了一个使用Python使用Tensorflow实现的网络。最后，我使用tf.train.Saver（）保存了模型。现在我想使用C ++通过这个经过预先训练的网络进行预测。

I've trained a network that I implemented with Tensorflow using Python. In the end, I saved the model with tf.train.Saver(). And now I would like to use C++ to make predictions using this pre trained network.

我该怎么做？有没有一种方法可以转换检查点，以便我可以将它与tiny-dnn或Tensorflow C ++一起使用？

How can I do that ? Is there a way to convert checkpoint so I can use it with tiny-dnn or Tensorflow C++ ?

欢迎任何想法：）谢谢！

Any idea is welcome :) thank you !

推荐答案

您可能应将模型导出到 SavedModel格式，它封装了计算图和保存的变量（ tf.train.Saver 仅保存变量，因此无论如何都必须保存图。）

You probably should export the model in the SavedModel format, which encapsulates the computational graph and the saved variables (tf.train.Saver only saves the variables, so you'd have to save the graph anyway).

然后可以使用 LoadSavedModel

You can then load the saved model in C++ using LoadSavedModel.

确切的调用取决于模型的输入和输出。但是Python代码看起来像这样：

The exact invocation would depend on what the inputs and outputs of your model are. But the Python code would look something like so:

# You'd adjust the arguments here according to your model
signature = tf.saved_model.signature_def_utils.predict_signature_def(                                                                        
  inputs={'image': input_tensor}, outputs={'scores': output_tensor})                                                                         


builder = tf.saved_model.builder.SavedModelBuilder('/tmp/my_saved_model')                                                                    

builder.add_meta_graph_and_variables(                                                                                                        
   sess=sess,                                                                                                                    
   tags=[tf.saved_model.tag_constants.SERVING],                                                                                             
   signature_def_map={                                                                                                       
 tf.saved_model.signature_constants.DEFAULT_SERVING_SIGNATURE_DEF_KEY:                                                                
        signature                                                                                                                        
})                                                                                                                                       

builder.save()

然后在C ++中您将执行以下操作：

And then in C++ you'd do something like this:

tensorflow::SavedModelBundle model;
auto status = tensorflow::LoadSavedModel(session_options, run_options, "/tmp/my_saved_model", {tensorflow::kSavedModelTagServe}, &model);
if (!status.ok()) {
   std::cerr << "Failed: " << status;
   return;
}
// At this point you can use model.session

（请注意，使用SavedModel格式还将允许您使用 TensorFlow服务来提供模型

(Note that using the SavedModel format will also allow you to serve models using TensorFlow Serving, if that makes sense for your application)

希望有帮助。

这篇关于使用Tensorflow检查点在C ++中还原模型的文章就介绍到这了，希望我们推荐的答案对大家有所帮助，也希望大家多多支持IT屋！