Convolutional neural network powered automation script for s0urce.io game

Features:

• Fully automated
• 0.999 network accuracy (but char detection could be better)
• Emulates direct input
• Leaves custom message to targets

Process:

Character detection:

Requirements:
Python 3.5.6
py-opencv 3.4.2
pillow 5.4.1
Tensorflow 1.10.0
Pynput
Keras 2.1.6

python main.py
You will have about 3 seconds to switch to browser with game. Target list, Target_info and cdm windows must be opened and not occlude each other. To stop script, press esc.

python brain.py
Process will take about 1 hour.

python record_dataset.py
You will have about 3 seconds to switch to browser with game.
cdm window must be opened and visible.
To collect data, just start playing the game, and the script will store keys and images from your screen. Script supports backspace key, which means you can correct your errors.
Also, script will validate input by compare lengths of detected word and your word.
Collected data will be stored in subfolder of dataset folder with name corresponding to char represented.
You must rerun script each time after finishing hacking the target.
NOTE: do not move the cursor using arrows - this will break all up.

Script was not tested in multi-monitor environment