With lighter
, focus on your deep learning experiments and forget about boilerplate through:
- Task-agnostic training logic already implemented for you (classification, segmentation, self-supervised, etc.)
- Configuration-based approach that will ensure that you can always reproduce your experiments and know what hyperparameters you used.
- Extremely simple integration of custom models, datasets, transforms, or any other components to your experiments.
lighter
stands on the shoulder of these two giants:
- MONAI Bundle - Configuration system. Similar to Hydra, but with additional features.
- PyTorch Lightning - Our
LighterSystem
is based on the PyTorch LightningLightningModule
and implements all the necessary training logic for you. Couple it with the PyTorch Lightning Trainer and you're good to go.
Simply put, lighter = config(trainer + system)
😇
Current release:
pip install project-lighter
Pre-release (up-to-date with the main branch):
pip install project-lighter --pre
For development:
make setup
make install # Install lighter via Poetry
make pre-commit-install # Set up the pre-commit hook for code formatting
poetry shell # Once installed, activate the poetry shell
Projects that use lighter
:
Project | Description |
---|---|
Foundation Models for Quantitative Imaging Biomarker Discovery in Cancer Imaging | A foundation model for lesions on CT scans that can be applied to down-stream tasks related to tumor radiomics, nodule classification, etc. |
If you find lighter
useful in your research or project, please consider citing it:
@software{lighter,
author = {Ibrahim Hadzic and
Suraj Pai and
Keno Bressem and
Hugo Aerts},
title = {Lighter},
publisher = {Zenodo},
doi = {10.5281/zenodo.8007711},
url = {https://doi.org/10.5281/zenodo.8007711}
}
We appreciate your support!