Ad pre-/postprocessing function to TensorFlow Serving Model

We are currently using tf2-yolov4 model with pre-trained weigths ([login to view URL]) to make inference from images. We compiled this to a tensorflow serving model (.pb) and made it availible with docker image and http REST API endpoint. This works well and we can send images to the REST API endpoint and get the predictions back. At the moment we need to pass in a numpy array in JSON format and get back the layers from the predictor. As we want to be able to send images in binary format to the REST API endpoint (for example using curl) we need a postprocessing or sheme conversion added to the exported model. As a result we want to have a well defined json output with class name mappings. We need help how to write the input/output processor to the existing model so we can use tensorflow serving to send binary data to the REST API endpoint and get back the results in the propper JSON format. This should be only a few line of codes and few modifications to save the model with the needed function but it can be assumed that we will have further tasks in the future for you.