YOLOv5 based Simple Fire Detector

1. Clone the Code

• Create anaconda environment

conda create -n fire_detector python=3.8
conda activate fire_detector

• Clone the code

git clone https://github.com/tiruss/Fire_Detector.git
cd Fire_Detector
git clone https://github.com/ultralytics/yolov5.git
cd yolov5
pip install -r requirements.txt
pip install pytube # for downloading youtube video

2. Download Dataset

• Download Fire Dataset [Download]

3. Convert XML format to TEXT format

• Run xml2yolo_text.py

python xml2yolo_text.py

• Then put the dataset into 'yolov5/data/fire_dataset'

4. Train the YOLOv5 Model

• Run train.py with proper dataset path
• You can change the hyperparameters in train.py [YOLOv5]

python train.py --data fire.yaml --cfg yolov5s.yaml --weights '' --batch-size 16 --epochs 100

• Track your training process on the tensorboard

tensorboard --logdir runs/train

5. Test Trained Network

• Download youtube video

python youtube_download.py

• Run detect.py

python detect.py --source [YOUR VIDEO] --weights [YOUR WEIGHT] --conf 0.4