Http Server

[spandansingh]

Added a dockerfile which exposed the model via rest api.

flask server will run on port 5005

URL - POST /query/{model} - model can be aesthetics or technical
Parameters - [
"<image_url_1>",
"<image_url_2>"
]

Response -

[
    {
        "image_id": "<image_name_1>",
        "mean_score_prediction": 4.780154329499055
    },
    {
        "image_id": "<image_name_2>",
        "mean_score_prediction": 4.780154329499055
    }
]

[clennan]

Hi, thanks a lot for the PR! It's a great idea to add a Flask server to the repo - there are a few things that need to be addressed in your PR:

1. Every time a request is send to an endpoint the respective model needs to be loaded which is inefficient. I would suggest to have only one prediction endpoint, in which the model only gets loaded once. You can achieve this e.g. with

def load_model(config):
    global model
    model = Nima(config['base_model_name'])
    model.build()
    model.nima_model.load_weights(config['weights_file'])
    model.nima_model._make_predict_function()  # https://github.com/keras-team/keras/issues/6462
    model.nima_model.summary()

...

if __name__ == '__main__':
    load_model(config)

    app.run(host='0.0.0.0', port=PORT)

The model should be provided as an argument when starting the Flask server. This has the downside that not both models are available in the same server, but makes it a lot more flexible for other users to serve their own models.

2. There are a lot of unneeded imports in src.server.py

3. Please add a description in the README of how to build the docker image and serve predictions

Many thanks!

[spandansingh]

Hi, thanks for considering it! Please have a look at the server.py changes and let me know if anything else needs to be done!

[Seluj78]

Up :D

[tombh]

It needed a few more bug fixes, but I'm using this now as well. Sorry not to push the fixes but I needed some personal customisations too. Anyway, thanks. Hope this gets merged soon!

[lucaspedrozaem]

Hi guys, thank you for the great work and sharing it !
Is there an exec file to make the Flask server run ?