This repository contains the source code for a reproduction/extension of the paper: KPRNet: Improving projection-based LiDAR semantic segmentation.
Autors:
Group 17:
Aden Westmaas (4825373)
Badr Essabri (5099412)
Guido Dumont (5655366)
Documentation:
Blog post: Project blog post
Original paper: KPRNet: Improving projection-based LiDAR semantic segmentation
Original GitHub: KPRNet github repository
Please follow the installation instruction of the original repository and add the packages listed in requirements.txt afterwards.
- Reproduction of the results in the paper
- Data augmentation to check robustness
- Implementation of the KITTI-360 dataset
Run run_inference.py in the source code in the original repository.
Run run_inference_data_augmentation.py to run the model with data augmentation. Within this file you can specify the amount of obscure used in the augmentation.
Run the Jupyter Notebook implementation_kitti360.ipynb.