Super-resolved Thermal Imagery for High-accuracy Facial Areas Detection and Analysis Alicja Kwasniewska, Jacek Ruminski, Maciej Szankin, Mariusz Kaczmarek
Supplementary materials for preprint submitted to Engineering Applications of Artificial Intelligence.
Please run
conda env create -f dresnet.yml
to create the conda environmnet with required packages.
Then run:
git submodule init/update
and link deeply-recursive-cnn-tf as deeply_recursive_cnn_tf using
ln -s deeply-recursive-cnn-tf deeply_recursive_cnn_tf
Training scripts are not currently available, please contact authors for assistance.
Current code supports inference using the provided DRESNet checkpoint trained on the collected thermal facial dataset using scale 2.
To run inference, use:
python test.py --model_name ckpts/ckpt_scale2/model_F96_D9_R3 --test_dir testing_images/
Due to privacy concerns dataset is not currently available, but please note that we are doing our best to provide it soon.
If you find this research useful or plan to use the provided model and checkpoints please cite:
@article{kwasniewska2020super,
title={Super-resolved thermal imagery for high-accuracy facial areas detection and analysis},
author={Kwasniewska, Alicja and Ruminski, Jacek and Szankin, Maciej and Kaczmarek, Mariusz},
journal={Engineering Applications of Artificial Intelligence},
volume={87},
pages={103263},
year={2020},
publisher={Elsevier}
}
@inproceedings{kwasniewska2019evaluating,
title={Evaluating accuracy of respiratory rate estimation from super resolved thermal imagery},
author={Kwasniewska, Alicja and Szankin, Maciej and Ruminski, Jacek and Kaczmarek, Mariusz},
booktitle={2019 41st Annual International Conference of the IEEE Engineering in Medicine and Biology Society (EMBC)},
pages={2744--2747},
year={2019},
organization={IEEE}
}
We'd like to thank authors of https://arxiv.org/abs/1511.04491 because DRESNet was created as an enhancement of their SR model DRCN and adapted specifically for thermal images.