CSGO-Aimbot-CNN

CS:GO Aimbot based on convolutional neural networks.

How it works

The program will capture 96x96 from the screen then it will
detect whether an opponent in the cropped frame. If there is it will
aim to his head.
The entire process will be displayed at your screen.
At the top left you will see how many frames the model process in a second (The higher the better).

GPU Support

22/10/2020 Update:
Download weights and extract in CSGO-Aimbot-CNN folder/assets.
GPU aimbot still under development but you can test it out now.
Support and tested on AMD RX480 and on GeForce GTX1050.
25/10/2020 Update:
After some testing although the efficence is way better using GPU.
The yolov3 converted to tensorflow is way less accurate than pure darknet weights.
So bad that with terrorist it barley works.
I suspect the accuracy get lost in the conversion. Currently I am looking
for a better conversion method to improve results.

How to make it work

1) You need to play on fullscreen windowed. For the overlays to show and for the frame capturing. 2) Mouse raw input needs to be disabled for the aimbot to work properly. 3) Resolution has to be 1920x1080. Otherwise you'll need to change screenWidth and screenHeight values.
Notice that fullscreen windowed mode just use native resolution. 4) Install all dependencies from reqiurements.txt file. 5) Open main.py and go to the game. Place an opponent in your FOV and your crosshair will be locked to his head.
For playing I would pick the aimbotOnly.py because the drawing slows the aimbot a little.
Tip:
Playing with raw input off may feel like mouse acceleration is on. You'll need to
disable in windows mouse setting the Enhance pointer precision option.

Keyboard Control

Compatibility issues

• In order to get a decent frames the model process in a second. I've only used
  the small fov for detecting opponents.
  You can tweak FOV but both values has to be the same.
  And be a multiple of 32.
  I found the limit to be 96x96. Getting 60 fps on my cpu!
  below that and the detections won't work.
  
• The higher the sensetivity the more snappy the aimbot is.
  After some testing I've found the middle ground to be 1.5 on 800dpi.
• The aimbot only tested on Window 10 with 1920x1080 resolution. But should work
  on other operating systems aswell.
• If you ran into No matching distribution found for tensorflow-directml
  trying to install tensorflow-directml. You need to install Python 3.7.7 Windows x86-64

  Bugs

  1) When more than one target in the FOV the aimbot locking to both target at same time.
  Even though we do calculate the distance between the target. If the target lost because
  the detection lost him. The aimbot will lock to the new closest target.
  I plan on solving this and improve performance at the same time using tracking method.
  I already gave a try to the existing tracking methods opencv provides. But
  they aren't good enough so I'm gonna implement my own.
  But for now I really lowered the FOV to avoid accidents.

Future plans

• [x] Adding support for aiming on terrorist opponents.
• [x] Adding support for sniper rifles
• [ ] Adding spray control that will target the head
• [x] Adding keyboard shortcut to toggle aimbot On / Off.
• [ ] Adding spray control crosshair.
• [ ] Switching between Sniper aimbot to rifle aimbot to pistol aimbot
  based on which weapon used.
  By detecting weapon text:
  
• [ ] Adding aim key
• [ ] Accelerating detection time using GPU (Currently in develop with DirectML to support AMD and NVIDIA)
• [ ] Improving results with Object Tracking

  Disclaimer

  This project is for educational purposes only.
  and it needs strong hardware to work efficiently.
  Meaning it won't work well for most common users.
  Developed in Python 3.7.7 Windows x86-64