Welcome to my GitHub repository. Here's a link to the video showcasing my project:
This repository contains the implementation of a road segmentation model using Convolutional Neural Networks (CNNs). This project aims to accurately segment road pixels in images, enabling autonomous vehicles to understand road boundaries.
Check out the demo video showcasing the road segmentation model in action.
Road segmentation is a critical task in computer vision for autonomous driving. This project demonstrates the implementation of a CNN-based road segmentation model that takes input images and predicts pixel-wise road and non-road classes.
We used the UC Berkeley BDD 100K dataset for training and evaluation. It contains labeled images with pixel-level annotations for road regions. The dataset includes a variety of scenarios, lighting conditions, and road types.
Fig. Test Image
Our road segmentation model is built using a FCN architecture, which has proven effective for image segmentation tasks. The architecture consists of an encoder that captures image features and a decoder that produces the segmentation mask. The model is trained end-to-end on the dataset.
-
Clone this repository:
git clone https://github.com/Lucifer2700/UCBerkley_Segmentation.git
-
Run the road segmentation on your own images:
python drivable-area-final.py
Our trained model achieves an accuracy of over 95% on the test dataset. The road segmentation output demonstrates accurate identification of road regions in various environments.
Here are some example results:
Fig. Model Prediction
Fig. Carla Prediction
Contributions to this project are welcome. If you find any issues or want to improve the model, feel free to submit a pull request or create an issue.
This project is licensed under the MIT License - see the LICENSE file for details.