Things different from original implementation of MultiMAE
- Aim to train with Python 3.10 on Kaggle
- Use
typed-argument-parser
instead ofargparse.ArgumentParser
=> Provide type hints, better for understanding code and ease of debugging.
Kaggle Notebook: Pretrain MultiMAE for RGB-D SOD
More details on Wandb report
This project is inspired by MultiMAE
@article{bachmann2022multimae,
title={MultiMAE: Multi-modal Multi-task Masked Autoencoders},
author={Bachmann, Roman and Mizrahi, David and Atanov, Andrei and Zamir, Amir},
journal={arXiv preprint arXiv:2204.01678},
year={2022}
}