In this project, we aim to explore and advance semantic segmentation for self-driving cars using three deep learning models: FCN, U-Net and DeepLab. Our study focuses on adapting and refining these models to enhance their accuracy and efficiency under diverse and challenging conditions, such as varying lighting and complex urban landscapes.
Group Members:
- Xinpeng Shan - shanxinp (x.shan@mail.utoronto.ca)
- Dechen Han - handeche (dechen.han@mail.utoronto.ca)
- Shi Tang - tangsh29 (tiffanyshi.tang@mail.utoronto.ca)
- Wendy Yusi Cheng - chengw54 (wendy.cheng@mail.utoronto.ca)
Code for data preprocessing, model training and testing can be found here:
https://www.kaggle.com/code/ekkkkh/fcn-u-net-deeplabv3-93b83a
Please run our project by clicking the Copy & Edit button on the top right.
Alternatively, we also saved the output in src_code_output.pdf and the source code in csc413-project-semantic-segmentation.ipynb.
Dataset: For training and testing our semantic segmentation models, we utilized two datasets.
The first dataset sourced from the Lyft-Udacity Challenge on Kaggle: https://www.kaggle.com/datasets/kumaresanmanickavelu/lyft-udacity-challenge.
The second dataset can also be found on Kaggle: https://www.kaggle.com/datasets/shivamaggarwal513/semantic-segmentation-car-driving/code.
Project Final Report: project_report.pdf