It is a sample application to detect a dog breed on the image. It consists of 3 parts:

• Create and train a model, and preparit for serving
• Detector API that provides a REST interface for the prediction
• Angular application that allows dog image selection and displays prediction results For details see my blog post From Keras model to Angular application

I created a CNN model for a dog breed prediction based on my learning project. I used Keras for the model implementation to simplify the things. Then I exported it and prepared to be served by a TensorFlow Serving server.

Since the TensorFlow serving talks gRPC I created a web service that provides an API via the REST interface. It is, actually, a facade that hides gRPC protocol with HTTP protocol, which is visible to the outside world.
The service is implemented as a NodeJS express application.

I created an Angular application that allows to select the images for prediction and displays the results. It is very simple and not nice, I just wanted to demonstrate how to build the pipeline from UI to the backend service and model hosted by TensorFlow Serving server.