DinoAI - Computer-vision based neural network training

Description

Dinoai is a neural network that is trained from the information provided only by image processing and nothing else. Usually, most neural networks are trained using data provided by internal variables that are pulled directly from the application's code itself. Dinoai is able to train itself purely from the data provided by the graphical output of the computer itself.

An eventual goal of the project is to apply the computer vision to video provided by camera input instead of direct input from the X Window System.

Project Organization

The cmake project was created using visual studio 2019 though the default cmake project. If interested, one is able to open the folder though Visual Studio 2019 and access a slew of streamlined tools such as automated copy and building to the server, syntax highlighting of cmake, automated cpp header importing, and many other features. While these features are nice, they are not required at all, and the project can be just as easily manipulated with any text editor such as vscode.

The project directory Dinoai contains the main entrypoint Dinoai.cpp and the two libraries Inputlib and Matchlib where Inputlib is a library for accessing bitmap information of the game though LibX11 and Matchlib is a library for image matching functions utlizing OpenCV.

Build

Installation and building

The project's Python scripts will run within a virtual python environment. First, create a launch virtual environment.

$ python3 -m venv --system-site-packages ./venv
$ source ./venv/bin/activate

To deactivate the virtual environemtn later (once you are done working with the project), run:

(venv) $ deactivate

Install the necessary packages

(venv) $ pip install --upgrade pip
(venv) $ pip install --upgrade tensorflow

Verify the installation of Tensorflow

(venv) $ python -c "import tensorflow as tf;print(tf.reduce_sum(tf.random.normal([1000, 1000])))"

Once the virtual environment has been configured, it can be reused in the future without repeating the installation steps.

The project can then be cloned onto the server and then built using make.

$ git clone "https://github.com/jqquanbeck/hpc-dino-game.git" # Clone code
$ cd hpc-dino-game                                            # Move into directory
$ make                                                        # Make the project

Running the code

The project can be run using the provided batching tool ./tools/run.sh from the project's root folder to run an array of instances for training.

$ ./tools/run.sh -N 15 # -N to decide the number of instances (one by default)