YOLOv8 Upgrade and AimBot

The detailed report in Report.pdf in the root dir

The slides for presentation is Presentation.pptx

Model config file and codes in ultralytics/cfg/models and ultralytics/nn, Shuffle Attention implemented in ultralytics/nn/modules/shuffleattention.py

parse_model function in tasks.py is changed to parse the new component

All the scale used is n which will make channels to 0.25 * channels

You can just modify the n inyolov8n-** in train files to use more powerful models which requires hardware supports, especially a strong enough GPU.

---

Requirements:

• opencv-python~=4.10.0.82
• torch~=2.3.0+cu121
• numpy~=1.26.4
• pandas~=2.1.4
• pywin32~=305
• einops~=0.8.0
• pathlib~=1.0.1
• pillow~=10.2.0
• requests~=2.31.0
• psutil~=5.9.0
• streamlit~=1.30.0
• matplotlib~=3.8.0
• tqdm~=4.65.0
• yaml~=0.2.5
• pyyaml~=6.0.1
• scipy~=1.11.4
• future~=0.18.3
• pygetwindow~=0.0.9
• bettercam~=1.0.0

---

Better to install pytorch through its website Pytorch to install the version correspond to your hardware

Use the commandline to install the requirements:

pip install -r requirements.txt

There are example models that were trained in models/*, you can use them directly

Original YOLOv8 Network

Train YOLOv8:

python3 train_yolov8.py

YOLOv8 with Shuffle-Attention (YOLOv8-SA)

YOLOv8-SA1:

Train YOLOv8-SA1:

Config the DATASET, epochs, batch in the corresponding train python file to custom the training.

python3 train_yolov8sa1.py

YOLOv8-SA3:

Train YOLOv8-SA3:

Config the DATASET, epochs, batch in the corresponding train python file to custom the training.

python3 train_yolov8sa3.py

YOLOv8 with 4 Detect Head (YOLOv8-4Detect)

Train YOLOv8-4Detect:

Config the DATASET, epochs, batch in the corresponding train python file to custom the training.

python3 train_yolov8_4detect.py

Evaluation

Config the eval.py first to specify the model and dataset(yaml path and the part like val or test)

Run the command:

python3 eval.py

to get the evaluation of the model.

AimBot (Applying the Model)

To train the model, use the dataset in dataset/data.yaml

• The data from Roboflow/Counter Strike 3

To apply the model, an AimBot for Counter Strike 2 was implemented

To use the AimBot, you firstly need to config the config.py file

The most important config is the model, which is the pt file path of the model you want to use

It is not recommend to change the SCAN_REGION_WIDTH and SCAN_REGION_HEIGHT which decide the region your model can see, unless you have powerful GPU that can support the calculation of larger size image.

CONFIDENCE_THRESHOLD is the least confidence of the result that are believed to be a target.

QUIT_KEY is the button to end the procedure, the default config is Q which means when Q is pressed during the procedure, it will exit immediately.

MODE is used to select the target you want to aim to.

MOUSE_MOVE_RATE is used to adjust your mouse rate, the lower means your mouse will move more slow but smoothly.

HEAD_SHOT_MODE is to set whether to aim the head of target

VISUAL is to set whether to show the view of model

After config complete, open your Counter Strike 2 and then start the procedure with:

python3 main.py

Wait for the procedure to start and use Caps Lock to control the aim

When Caps Lock is on, the procedure will aim to the target as config

When Caps Lock is off, the procedure will just view the region but will not move your mouse

Use QUIT_KEY (original set to Q) at any time to exit the procedure.