We have released our dataset proposed in paper 'Object Detection in Hyperspectral Images'. Raw hyperspectral images and processed data (96-channel) can be found at [baidu cloud, password: 6shr], [Onedrive].
1.Main libraries
torch==1.1.0
cuda==10.0
libtiff==0.4.2
2.Training
generate label json file
python create_data_lists.py
then
python train.py
Note that due to samll scale of training dataset, the mAP may have relatively large jitters (about 2 mAP) with different random seeds.
3.Eval
python eval.py
our pretrained model: baidu cloud (password:mg57)
4.Acknowledgements
Our work is implemented based on this repo, thanks for this work.
If you use our work in your researches, please cite our paper as follow:
@article{yan2021object,
title={Object Detection in Hyperspectral Images},
author={Yan, Longbin and Zhao, Min and Wang, Xiuheng and Zhang, Yuge and Chen, Jie},
journal={IEEE Signal Processing Letters},
volume={28},
pages={508--512},
year={2021},
publisher={IEEE}
}