General

• This is the code we used to participate track 5 (Detecting Violation of Helmet Rule for Motorcyclists) of AI CITY CHALLANGE 2023.
• We used a YOLOv8 for detector and strongsort for tracker.
• For details about dataset, you should go to the link above.

Requirement

• For visualizing-web: pip install -r visualizing-web/requirements.txt
• For yolov8, refer this repo: https://github.com/ultralytics/ultralytics
• For strongsort, refer this repo: https://github.com/kadirnar/strongsort-pip

Visualizing data

• If you want to visualize video, move videos to 'visualizing-web/static/data/bbox-video'
• Then, run the following command: python visualizing-web/app.py

Training

• To train yolov8, refer this repo: https://github.com/ultralytics/ultralytics
• To train feature extractor of strongsort, refer this repo: https://github.com/kadirnar/strongsort-pip

Inference

• cd to yolov8 folder: cd yolov8
• We provide the checkpoint file of YOLOv8 on this dataset: https://drive.google.com/file/d/1P-VjBKllb4KcqoNTYQsXhTYOMwmkmu9X/view?usp=sharing
• Change the path folder of videos and path to checkpoint file of YOLOv8 in inferece.py if necessary, then: python inference.py
• However, there will be some bounding boxes predicted which extend outside the image. We need to modify the output with the following command: python post-preprocessing.py