Skip to content

mrbid/CSGO_TENSOR_TRIGGER

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 

Repository files navigation

CSGO_TENSOR_TRIGGER

A series of machine learning trigger bots for Counter-Strike: Global Offensive (CS:GO).

I made this from the ground up, I collected the dataset of 17,440 samples myself, initially manually and then assisted with automation. The initial working prototypes I created from the ground up in C, now the final solution uses Tensorflow Keras. There's a series of articles I wrote while I was working on those original protoypes that you can read here.

Seems this dataset also works to some extent on CSS and 1.6 / CZ, to be honest it will probably have similar results on any game with humanoid characters.

GOBOT12 Video: https://youtu.be/R-nCL5zqZBQ
GOBOT11 Video: https://youtu.be/UMBqk8CAe04

The default builds in PRED_CNN and PRED_FNN use ONNXRUNTIME and perform at around 500 FPS.

  • A Keras2c build has been added here. [220-250 FPS]
  • A libonnx build of the Keras CNN is here. [130-150 FPS]
  • A TBVGG3 CNN version is here. [330 FPS] (smaller less accurate model)
  • A standalone version of the FNN is here. [450 FPS] (less less accurate than a CNN)

Crosshair

The game crosshair will get in the way of the bots ability to detect player models on the screen, you have a few options to mitigate this, each option is just as effective as the other:

  • Change your crosshair settings in CS:GO to a single green pixel center dot with no outline.
  • Set your CS:GO crosshair to invisible by enabling the alpha channel and sliding it to invisible.
    • Then use the crosshair provided by the bot which encapsulates the scan region with a square.
    • Or if your monitor supplies a built-in crosshair, use that. (because it does not get written to the game renders, it's a hardware overlay)

Tips 'n Tricks

The bot generally regulates your rate of fire for you but if you need more controlled bursts enable sample capture, this will ensure shots are fired in single rhythmic bursts. (it's a side effect from ensuring duplicate samples of the same frame are not captured)

Information

This repository holds the best releases from a series of articles I made that document my research into making a CS:GO auto-trigger bot using machine learning: https://james-william-fletcher.medium.com/list/fps-machine-learning-autoshoot-bot-for-csgo-100153576e93

Latest datasets are now maintained over at TFNN/DOCS/DATASETS.