Skip to content

Burned area segmentation from Sentinel-2 using multi-task learning

License

Notifications You must be signed in to change notification settings

links-ads/maclean-burned-area-segmentation

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

67 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Robust Burned Area Delineation through Multitask Learning

Code and dataset for Robust Burned Area Delineation through Multitask Learning, (ECML PKDD, MACLEAN Workshop 2023)

Note

Dataset available at hf.co/datasets/links-ads/wildfires-cems.

Installation

First, create a python environment (preferably Python 3.10), then install the package itself:

$ python -m venv .venv
$ source .venv/bin/activate
$ pip install -e .

Then, install mmsegmentation:

$ pip install -U openmim
$ mim install mmengine
$ mim install "mmcv>=2.0.0"

At last, you can install mmseg by running:

$ pip install "mmsegmentation>=1.0.0"

Data preparation

Make sure the dataset is formatted following the schema in datasets.py. By default, the script expects data to be located in data/ems. This can also be achieved by symlinking the dataset to this location:

$ ln -s <absolute_path_to_data> data/ems

Experiments

The experiments exploit a mixture of PyTorch Lightning's and mmsegmentation's logging features to handle most of the experiment's configuration. Each run creates by default a folder in output/ with the following structure:

outputs/
└── <experiment_name>/
    ├── weights/
    │   ├── <checkpoint_1>.pth
    │   ├── <checkpoint_2>.pth
    │   └── ...
    ├── config.py
    └── tensorboard logs...

This folder can later be used to resume training or perform inference.

Training

Using mmseg configuration files (sort of), you can train a model by running:

$ python tools/launch.py train -c <config_path>

Testing and Inference

Once a model has been trained, you can test it by running:

# if you just want to compute the metrics
$ python tools/launch.py test -e <experiment_path> [-c <checkpoint_path>]
# if you want to save the predictions
$ python tools/launch.py test -e <experiment_path> [-c <checkpoint_path>] --predict

Where experiment_path is the full or relative path to the experiment directory (including the version subdir), and checkpoint_path is the full or relative path to the checkpoint file (including the .pth extension).

Citing this work

@inproceedings{arnaudo2023burned,
  title={Robust Burned Area Delineation through Multitask Learning},
  author={Arnaudo, Edoardo and Barco, Luca and Merlo, Matteo and Rossi, Claudio},
  booktitle={Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
  year={2023}
}

About

Burned area segmentation from Sentinel-2 using multi-task learning

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published