This repository contains the code for the winning model of the Deep Learning Hackathon focused on image segmentation for autonomous vehicles using supervised learning techniques. Our model leverages state-of-the-art deep learning architectures to achieve high accuracy and efficiency in segmenting images critical for autonomous driving.
Problem Statement
Background
Autonomous vehicles (AVs) are revolutionizing transportation by promising safer, more efficient, and more convenient travel. A critical component of AV technology is the ability to accurately perceive and understand the surrounding environment in real-time. Image segmentation plays a pivotal role in this perception by enabling AVs to identify and distinguish various elements on the road, such as lanes, vehicles, pedestrians, traffic signs, and obstacles.
Objective
Design and train an image segmentation model on an autonomous vehicle dataset. The model must be able to segment the scene in images into 13 classes, including lanes, vehicles, pedestrians, traffic signs, and obstacles. The input size of the image should have a height and width of 256x256 with 3 channels.
The dataset used for this hackathon includes labeled images from autonomous vehicle environments, ensuring a robust and diverse training set. The link to the dataset: https://www.kaggle.com/datasets/bishwashk/iims-hackathon-2024