create model for inception model

[R-Miner]

Hello Vetal1977
In program, svnh_semi_supervised_model_saved.py, at line 71 you create the GAN model. Similarly how can I create an InceptionV3 model Keras model?
I am trying to adapt your code to serve an inception model.

Thanks
Rminer

[vbezgachev]

Hi @R-Miner

Sorry for some delay. I didn't serve a Keras model with TensorFlow until now. If I need, I would covert a Keras model into TensorFlow estimator and serve it then.

Vitaly

[R-Miner]

Thanks Vitlay for looking into the matter. Will update you about how it went.

I did serve a sequential Keras mode successfully by adding a input layer which accepts a string and does the preprocessing in an additional layer just before exporting the model. But this is an overhead to the existing model and would take more computational time if the model is really deep and also very inefficient. That's why I again started looking for simpler solutions.

Your suggestion, Keras to estimator would serve the purpose I assume. Thanks again!!

I would like to revise my question in detail. I can successfully deploy any Keras/tensorflow model in an Ubuntu Linux machine. Now my aim is to run the client in a windows machine with no python installed and developed a C# client which sends an input string tensor to the server. But in general inception model or any simple Keras model by default accepts an image as a float for classification. So I am trying to add a layer or prepend another model which accepts a string input image tensor and then process the string tensor to float tensor and feed it into the existing model and export the new one that way the model is ready to accept a string tensor send from client. So hopefully by converting the model to estimator serves the purpose.