Drivable area Segmentation using BDK100 Dataset with UNet

Table of Contents

• Project Overview
• Dataset
• Model Architecture

Project Overview

This project focuses on drivable area detection using the BDK100 dataset and a UNet model. The goal is to accurately identify lane lines which you can drive in, which is a critical task for autonomous driving systems.

Dataset

Click here for dataset description and download

Model Architecture

I have implemented a UNet model for drivable area segmentation. The UNet architecture consists of an encoder-decoder structure that helps in capturing both local and global information for precise segmentation.

Key Features:

• Encoder: Custom CNN model was used here as in the paper
• Decoder: Upsampling with skip connections
• Loss function: Dice loss
• Optimizer: Adam

to run this project locally, download the utils.py file, which has the required utils to run the UNet_code