Car Brand Detection using YOLOv4

This project is developed to detect car brands in images and videos using YOLOv4. The model is trained with 10 car brands' images using pretrained YOLOv4 model.

Change path_name variable in pyhton files to experiment with images and videos.

To run:

• pip3 install -r requirements.txt

• Download the model weights and put them in weights folder.

• To generate a car brand detection image on data/chrysler.jpg:

  python yolo_opencv.py
  

  A new image chrysler_yolo4.jpg will appear which has the bounding boxes of the cars in the image.

• To read from a video file and make predictions on data/mercedes.mp4:

  python vid_yolo_opencv.py
  

  This will start detecting car brands in that video, in the end, it'll save the resulting video to output_video/mercedes.avi

Class Names

Following car brands are used to detect in this project.

• Audi
• BMW
• Bentley
• Chrysler
• Ford
• Honda
• Hyundai
• Mercedes-Benz
• Nissan
• Toyota

Dataset

The dataset is reconstructed from the Stanford AI Lab - Cars Dataset by preprocessing properly to make it convertible to YOLO format.

Technical Details

This project is developed using darknet framework and conda environment with following hardware and software configurations:

• GPU - Nvidia GeForce GTX 1660 Ti 6 GB
• CUDA - v11.3
• cudnn - v8.2.0
• OPENCV - v4.5.1

Demo

Watch the demo below.