The purpose of this project is pretty simple, and was a pretty good first look into AI. Detect if there is an enemy on the crosshair with an image neural network, and then flick away.

The training data was sourced from video of me running around the Dust 2 map. I then cut the frames in Premiere and programmatically labeled the video frames as with or without a target. This was then fed into a Tensorflow/Keras network of 3 convolutional 2D layers, which then output whether there was someone in the crosshairs. The output was quite effective, although I didn't do a full test sequence, and the gameplay results are in the YouTube video above.