/SMTL

The implementation of "Safe Multi-Task Learning".

Primary LanguagePython

Safe Multi-Task Learning

Prerequisites:

  • Python3
  • PyTorch ==1.6.0 (with suitable CUDA and CuDNN version)
  • torchvision == 0.7.0
  • Numpy
  • argparse
  • transformers == 4.6.1

Training:

You can run "cityscapes/run.sh" to train and evaluate on the CityScapes dataset.

You can run "nyu/run.sh" to train and evaluate on the NYUv2 dataset.

You can run "pascal/run.sh" to train and evaluate on the PASCAL-Context dataset.

You can run "taskonomy/run.sh" to train and evaluate on the Taskonomy dataset.

Citation

If you use this code for your research, please consider citing:

@article{guo2021safe,
  title={Safe Multi-Task Learning},
  author={Guo, Pengxin and Ye, Feiyang and Zhang, Yu},
  journal={arXiv preprint arXiv:2111.10601},
  year={2021}
}

Contact

If you have any problem about our code, feel free to contact 12032913@mail.sustech.edu.cn.