Skip to content

🌲 Aimbot powered by real-time object detection with neural networks, GPU accelerated with Nvidia. Optimized for use with CS:GO.

License

Notifications You must be signed in to change notification settings

petercunha/Pine

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

43 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Pine

Pine is an aimbot powered by real-time object detection with neural networks.

This software can be tweaked to work smoothly in CS:GO, Fortnite, and Overwatch. Pine also has built-in support for Nvidia's CUDA toolkit and is optimized to achieve extremely high object-detection FPS. It is GPU accelerated and blazingly fast.

Demo

Demo video

Pine Demo GIF=

Installation

  • Install OpenCV with Python3 bindings
  • Install any missing dependencies with Pip
  • Install pywin32, the Windows API bindings for Python.
  • Install the Nvidia CUDA toolkit if you have a compatible GPU
    • If you don't have an Nvidia GPU, Pine will try to use OpenCL instead of CUDA
  • Also for acurate long Quick Mouse Movements please Disable Mouse acceleration. Thanks!
  • Then run the program with Python3.

$ python3 pine.py

What games does it work with?

This release is currently optimized for CS:GO, but I plan adding more game configs in the future

Why Neural Networks?

Well, neural network aimbots are great for a lot of reasons... Probably most importantly, they never access the memory of the game, so they are practically invisible to "Anti-Cheat" sofware. Additionally, they can abstract their capabilities to many different games without code modifications. The only issue:

Neural Network Aimbots have always had one big problem: their target detection is okay, but their inference time is terrible. Even HAAR Cascades are a bad fit, since they have decent speed but horrible accuracy. MobileNetSSD and Faster R-CNN were sorta ok if you had a Nvidia Titan X with CUDA drivers. But let's be honest, who the hell can afford a Titan?

Enter YOLOv3, tiny edition. Detection scores are decent, and inference times are... WHAT? 220 FPS WITH A 33% mAP?! For reference of how absolutely insane this is, SSD513 gets about 8 FPS with 50% mAP.

YOLO Network Speed Graph

Finally, a real-time object-detection network that will run on my dinky AMD M370X from 2015! This network is what inspired me to build Pine.

Architecture

YOLOv3: The Internal Object-Detection Engine

At its core, Pine's internal object detection system relies on a modified version of the YOLOv3 neural network architecture. The detection network is trained on a combination of video game images and the COCO dataset. It is optimized to recognize human-like objects as quickly as possible. The detection engine can be abstracted to many FPS games, including CS:GO, TF2, and more.

YOLO Neural Network

OpenCV: GPU-Enabled Image Processing

OpenCV is at the heart of Pine's image processing capabilities. Not only does it provide an abstraction layer for working with the screen capture data, but it it also allows us to harness the power of GPU hardware by interfacing CUDA and OpenCL. After Pine takes a capture of the user's screen, it uses OpenCV to process that image into a form that is recognizable to the object-detection engine.

OpenCV Image Processing Architecture

Special Thanks

Thanks to everyone who made this project possible. I'd like to give special shoutouts to the following people:

  • FidgetySo for his contributions to the codebase. He leveraged native Windows APIs to improve the performance of Pine by up to 300%.
  • Adrian Rosebrock, PhD, for his invaluable resources on machine learning. Check out his blog: PyImageSearch