Our project is developed based on mmsegmentation. Please follow the official docs for installation and dataset preparation.
Here is a full script for setting up with conda.
conda update --all
conda install pytorch=1.7.0 torchvision torchaudio cudatoolkit=11.0 -c pytorch
pip install mmcv-full==1.2.7 -f https://download.openmmlab.com/mmcv/dist/cu110/torch1.7.0/index.html
pip install -e . # or "python setup.py develop"
pip install -r requirements/optional.txt
cd ../..
cd project/Swin-Transformer-MulT
pip install -e . # or "python setup.py develop"
pip install -r requirements/optional.txt
cd ..
sudo apt update
sudo apt install libgl1-mesa-glx # Originally posted by @rsignell-usgs in https://github.com/conda-forge/pygridgen-feedstock/issues/10#issuecomment-365914605
cd SETR
# Training
# bash ./tools/dist_train.sh ${config} ${gpu_num}
bash ./tools/dist_train.sh configs/METR/MMETR_40k_NYU_bs_8.py 2
# Evaluation
# bash ./tools/dist_test.sh ${config} ${pth_file} ${gpu_num} --eval mIoU
bash ./tools/dist_test.sh configs/METR/MMETR_40k_NYU_bs_8.py /metr.pth 2 --eval mIoU
cd Swin-Transformer-MulT
# Training
# bash ./tools/dist_train.sh ${config} ${gpu_num}
bash ./tools/dist_train.sh configs/swind/swind_dea_large_patch4_window7_40k_NYU.py 2
# Evaluation
# bash ./tools/dist_test.sh ${config} ${pth_file} ${gpu_num} --eval mIoU
bash ./tools/dist_test.sh configs/swind/swind_dea_large_patch4_window7_40k_NYU.py <path>/dea4.pth 2 --eval mIoU
In the example above, model decode_head type is set to SwinTransformerDEA4
.
@InProceedings{Bhattacharjee_2022_CVPR,
author = {Bhattacharjee, Deblina and Zhang, Tong and S\"usstrunk, Sabine and Salzmann, Mathieu},
title = {MulT: An End-to-End Multitask Learning Transformer},
booktitle = {Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2022},
pages = {12031-12041}
}
License: Creative Commons Attribution Non-commercial No Derivatives