only can do a nice performence in killing bot
this project is base on yolov5
most of my tests are run in windowed fullscreen csgo
the main1.py in mian directory can work like the video blow
using these commands in game:
sv_cheats 1
weapon_accuracy_nospread 1
weapon_recoil_scale 0
otherwise the speed is too fast for recoil controling
wx_camera_1666798546016.mp4
compare to human aiming from bilibili-大东彦
https://www.bilibili.com/video/BV1nU4y1G73M
wx_camera_1666682766159.mp4
the ghub_device.dll is used to control keyboard and mouse by logitech driver, so it works when raw-input on
ghub_device.dll require logitech ghub version below 2021.6
the video above was using a model i trained from 600+ images with 350 epoch
screenshot.py can take a screenshot when press the second mouse side key
to get image to label
roboflow is my labeling workspace https://universe.roboflow.com/lz-k-eigsu/csgo-ql8ba now it got almost 1000 pictures
training model use yolov5s
-
main1.py will auto aim and shot if mouse side botton pressed, and press again can stop
-
main2.py is my attempt to realize full auto afk in deathmatch (failed)
i throught locate enemies by voice, but this is far beyound my capacity, using +left instaed
in game command
bind "l" "+left"
- main3.py trying to use threads to sperate detect&aim function and mouse auto-click funtion with hotkeys (pynput's keyboard and mouse listener may conflict, fail)
there's a problem when i using threading in mouse control func
the mouse control thread is a loop, and when i add a time.sleep(0.1) into it, it will work normally
but when change the time of sleep less than 0.1, the mouse will flow about
when i delet time.sleep(), it will randomly move to two pole very quick
still unsloved
- main4.py let detect work from beginning to the end, using hotkey to control autoaim switch and +left
install requirment just the same like yolov5 offical tutorial
also need to install cudatoolkit, tensorrt, pynput,
tensoRT installing follow offical page https://docs.nvidia.com/deeplearning/tensorrt/quick-start-guide/index.html#install
conda install cudatoolkit
pip install pynput
the codes isn't very long, they are all less than 200 lines, so i don't have too many things to explain
im a beginner in coding, a noob actuaclly. the learing curve last for about 2 mouthes, from writing a list in python. so those codes are immatrue obviously, and my english is still bad 🤣
i do this just because interest, than the interest became some kinds of wild ambition. i was always thinking of how to make this project batter, so i used tensorRT, learned deepsort try to track characters in game. the last thing failed because i can't trans grabed images to a video stream as a input. https://github.com/mikel-brostrom/Yolov5_StrongSORT_OSNet
millions of thanks to every opensource community
millions of thanks to yolo develpors
millions of thanks to csdn and stack overflow for sloving many weird problems
millions of thanks to my pc
screenshot from my pc when train the last model
