Traffic Light Detection with YOLOv5

This repository contains the implementation of a traffic light detection model fine-tuned using YOLOv5 on a custom dataset. The dataset is provided by Roboflow.

Table of Contents

• Introduction
• Dataset
• Setup
• Training
• Results

Introduction

Traffic light detection is a critical component in autonomous driving systems. This project aims to fine-tune a pre-trained YOLOv5 model to accurately detect traffic lights in images. Paper

Dataset

The dataset used for this project is available on Roboflow and includes images and annotations for traffic lights. You can download the dataset from here.

Setup

To get started, clone this repository and install the necessary dependencies:

notebook

git clone https://github.com/ultralytics/yolov5.git

and split train-valid dataset

Training

Use the following command to start training:

python train.py --img 360 --batch 256 --epochs 50 --data ../data.yaml --cfg models/yolov5s.yaml --weights yolov5s.pt --name traffic_light_yolov5s_results --entity taewan2002 --project traffic_light_yolov5s

Results

wandb link

Test modal with this camera

Convert the model

Apply the model

Test the model with camera

It works well!!

More Work

How about vision-language model (VLM), PaliGemma-3b link