Extraction test

The test_game.py script provide visual qualitative evaluations of the selected method. It will run the game for some steps, then plot a frame with squares at the bouding boxes of the extracted objects.

The parameters are:

  • -g GAME, --game GAME : game to evaluate (e.g. ‘Pong’)

  • -p PATH, --path PATH : the path to a potential DQN agents’ model

  • -i INTERVAL, --interval INTERVAL : The frame interval (default 10)

  • -m {vision,revised}, --mode {vision,revised} : The extraction mode

  • -hud, --hud : If true, also detects objects from the HUD

We provide hereafter an example on the Pong game using the vision method.

A picture of the vision extraction method on Pong.