ma1-ml-proj2's Introduction

Project Road Segmentation

Git LFS

This project uses Git LFS. Make sure to install it before cloning the repository. See Git LFS. If the bandwidth is exceeded, you can use the mirror on Gitlab: gitlab.com/Maeeen/ma1-ml-proj2.

Project architecture

The project is organized as follows

/checkpoints                # Saved model checkpoints for different architectures
/data                       # Train and test data
/notebooks                  # Past experiments
/src/road_segmentation      # Source code
  /models                   # Models definitions
  /utils                    # Utility functions
  mask_to_submission.py     # Somes functions related to submissions
  submission_to_mask.py     # Somes functions related to submissions
/submissions                # Generated submission files for AIcrowd
infos.md                    # Submissions and checkpoints descriptions
main.ipynb                  # Training code
predict.py                  # Generate predictions on the test set using a trained model. Using Unet by default.

Setting up the environment

First, make sure to clone the repository and to place you on the main branch.

Conda

Dependencies can be managed by conda in the following way. This ensures reproducibility of the environment at the libraries level.

conda env create -f environment.yml
conda activate road-segmentation

# Install pytorch, torchvision, torchaudio, torchinfo

conda install -c conda-forge torchinfo

pip install -e .

Note: make sure to install pytorch with the right version for your system. You can follow the instructions here.