shibuiwilliam / e2e_tensorflow_serving Goto Github PK

Nice job William, however I tried to figure out a detail problem.

Here, for example, line 8 in build_inceptionv3.py:

def get_label() -> List[str]: 
    with open("./ImageNetLabels.json", "r") as f: 
        labels = json.load(f) 
    return labels

The labels are loaded from a external JSON file when the model class is initialized, and extra processings are added through signatures, that's alright.

But here is my question, it seems that the path of the label JSON file is never attached as a volume parameter of tf-serving docker container, how does it works in the post-processing step? Is the label info added together with the model weights located in the saved_model/inception_v3 when you save your model?