object-detection

This repository contains an implementation of object detection using YOLOv8 specifically designed for detecting weapons in images and videos. The repository includes pre-trained models and the python code to use your webcam for testing.

Resources Used

Language: Python 3.9

Dataset: https://universe.roboflow.com/fend-tech/weapon-detection-dinou

Here I have used a labelled dataset but if are unable to find a pre-labelled dataset you can label it yourself.

Manually labelling your dataset

Split your dataset into three classes train test valid.

Use labelImg -> https://github.com/heartexlabs/labelImg Make sure to select the txt format for the labels as that is what YOLO requires.

YOLO also requires a data.yaml file. If this file is not generated by labelimg simply create one, here is what my data.yaml contains :

names:
- Grenade
- Gun
- Knife
- Pistol
- handgun
- rifle
nc: 6
test: ../test/images
train: ../train/images
val: ../valid/images

Setting the Data Systematically

The Directory should be as follows:

object_detection
             └── train
                   ├── images
                   │      ├── image_1.jpg
                   │      ├── image_2.jpg
                   │      └── ...
                   │
                   └── labels
                          ├── image_1.txt
                          ├── image_2.txt
                          └── ...
             └── test
                   ├── images
                   │      ├── image_1.jpg
                   │      ├── image_2.jpg
                   │      └── ...
                   │
                   └── labels
                          ├── image_1.txt
                          ├── image_2.txt
                          └── ...
             └── valid
                   ├── images
                   │      ├── image_1.jpg
                   │      ├── image_2.jpg
                   │      └── ...
                   │
                   └── labels
                          ├── image_1.txt
                          ├── image_2.txt
                          └── ...
             ├── data.yaml
             ├──weapon-detection.ipynb
                

Steps to train the model on your custom dataset :

- Create an environment

$python -m venv venv

- Activate Environment

Windows

# In cmd.exe
venv\Scripts\activate.bat
# In PowerShell
venv\Scripts\Activate.ps1

Linux

source myvenv/bin/activate

- Install YOLOv8 via the ultralytics pip package

$pip install ultralytics

- Test your YOLO installation

$yolo task=detect mode=predict model=yolov8n.pt conf=0.25 source='https://media.roboflow.com/notebooks/examples/dog.jpeg'

- Train YOLOv8 on a custom dataset

$yolo task=detect mode=train model=yolov8s.pt data={dataset location}/data.yaml epochs=100 imgsz=640

Your model will begin training and run for several minutes, or hours, depending on how big the dataset is and which training options you chose. A folder named runs will be created.

- Validate with a new model

yolo task=detect mode=val model=/runs/detect/train/weights/best.pt data={dataset location}/data.yaml

- Predict with a custom model

yolo task=detect mode=predict model={HOME}/runs/detect/train/weights/best.pt conf=0.25 source={path to the image u wish to run inference on}

You can also run prediction on an entire folder containing test images

yolo task=detect mode=predict model={HOME}/runs/detect/train/weights/best.pt conf=0.25 source={dataset.location}/test/images

My Results

Test set

Confusion Matrix

Loss over time

Note :Roboflow is a quick way to get your hands on some labelled data. https://universe.roboflow.com/roboflow-100