Automation of "tents & trees" Game

Note: you will need to manually create private/ directory.

Requirement:

• an Android phone.
• opencv-python and python 3.7. I have no idea nor do I care why it only works on 3.7.
• adb, we will do screenshots and issue taps using it.
• a compiled binary of tents-demo.
• android_input_agent

Here are few steps to have a working program, if my memory serves.

• Collect sample for different sizes.

  Each different puzzle size needs one. Collect and store them under private/samples/{h}x{w}/, where h is the screen height of the phone and w screen width. The naming pattern is sample-{size}x{size}.png where {size} is the size of the puzzle (a number from 5 to 22).

  To take phone screenshots, use adb exec-out screencap -p > sample-{size}x{size}.png.

• Generate private/preset.json.

  This file allows us to quickly find cell positions for all different sizes of the puzzle board.

  For this you just need to run cd py; ./gen_preset.py.

• Collect digit samples.

  We'll need to visit through those samples again in order to collect digit samples for recognition to work. Tagging with cd py; ./tagging.py.

  For the tagging process, it is recommended that you tag at most one sample for each digit at one iteration, this allows us to get a smaller set of samples (untagged samples will be removed if next time it can be recognized (by newly added samples) with a good score.

• Have a working tents-demo binary.

• Set PYTHONPATH so that it points to a local copy of android_input_agent's client

• Set environment variable TENTS_DEMO_BIN to the location of the binary.

• Set environment variable AIA_PORT, which should point to a running server of android_input_agent.

• cd py/ then ./solver.py when the phone is at a game screen.

• (Optional) Set environment variable PUZZLE_RECORDS to a file path to append recognized puzzles to it.

• cd py/; ./analyze_samples.py can used to gather some analysis, this is mostly just for experimenting with threshold methods.