Servision

Servision is a modular API server that can process media with machine learning and computer vision algorithms

How to Use

• Servision can be used to run algorithms on user-uploaded files. The algorithms can be anything provided a few simple requirements are met:

  • Its module name is added to the 'algorithms' part of the config.json

  • All required imports are in 'requirements.txt'

  • It's in its own module '{name}.py' in the server/algorithms folder

  • The algorithm module contains the function proc_call(token, store_path, out_path) -> tuple[int, str]

    • token: The task id, {algo}_{uid}_{timestamp}, which should be used in any output filenames
    • store_path: the location of the input file
    • out_path: the folder where any output files should be written
    • returns (int, str): Status code (0=ok, 1=error), and message, which if status=0, returns out_path

    It is recommended that error checking be done in proc_call(), i.e. checking if the paths exist

• Servision is integrated with firebase for authentication.
  • Generate a new private key and download the certificate from your firebase console
  • Add the path to your certificate as the environment variable FB_CERT_PATH
  • When making requests to the server, include the headers authorization with a valid token.

Tasks left

• Fully test firebase integration, apply user data to jobs and use auth checks on all queries
• Create a sample model that is computationally intensive and will produce multiple files,
  • Then test the task check and download queries
  • Benchmark performance by queueing multiple tasks