Radar-Detection with Yolov8

Self-Driving Cars Project for NYCU Courses, Fall 2023

Objectives

Used existing radar images to train a detector with the goal of recognizing various vehicles in the images (e.g., trucks, minibuses, bicycles, buses)

Bonus

Verified model performance using data from Guangfu Road, Hsinchu, Taiwan

Complete File Structure - vehicle_detection

├── bonus_data
│   ├── images
│   │   ├── train
│   │   └──  val
│   ├── labels
│   │   ├── train
│   │   └──  val
│   └── train_data # train_data from google drive bonus folder
│       └── ...
├── runs
│   └── dectect # weights will be saved here
├── ultralytics # tools for yolo
│   └── ...
├── yolo_best # best weights and json for competition and bonus
│   ├── yolov8l_bonus_train6_best_pred.json
│   ├── yolov8l_bonus.pt
│   ├── yolov8m_.json
│   └── yolov8m_.pt
├── yolov8_dataset
│   ├── images
│   │   ├── train
│   │   └──  val
│   └── labels
│       ├── train
│       └──  val
├── all_eva #Evaluation code
├── data_process_bonus.ipynb # generate labels for traning bonus model
├── data_process.ipynb # generate labels for traning competition model
├── dataset.yaml
├── README.md
├── weights_to_json.py
├── yolov8l.pt
├── yolov8m.pt
├── yolov8n.pt
├── yolov8s.pt
└── yolov8x.pt

Yolov8m Hyperparameters and Configurations

The best model performance was achieved using YOLOv8m.

parameters	default value	optimizer value
epochs	100	65
optimizer	auto	Adam
learning rate(lr0)	0.01	0.0001
patience	50	25
augmentation	True	True
batch size	16	16

Result Demo

video : https://youtu.be/BmGU1UOnfzY

License

This project is released under the MIT License.