This is the official implementation of the paper:
J.-H. Park, W.-J Nam and S.-W Lee, "A Two-Stream Symmetric Network with Bidirectional Ensemble for Aerial Image Matching," Remote Sens., 2020, Vol. 12, No. 6, pp. 465
[Journal][arXiv]
- Python 3
- PyTorch, torchvision
- pretrainedmodels
- scikit-image, pandas, opencv
- termcolor, tqdm
- googledrivedownloader
- demo.py demonstrates the results on the samples of aerial image dataset
- train.py is the main training script
- eval_pck.py evaluates on the aerial image dataset
Note that, models must be downloaded to the 'trained_models' folder.
| Backbone Network | PCK (tau=0.05) | PCK (tau=0.03) | PCK (tau=0.01) | Download Link |
|---|---|---|---|---|
| ResNet101 | 93.8 % | 82.5 % | 35.1 % | [here] |
| ResNeXt101 | 94.6 % | 85.9 % | 43.2 % | [here] |
| Densenet169 | 95.6 % | 88.4 % | 44.0 % | [here] |
| SE-ResNeXt101 | 97.1 % | 91.1 % | 48.0 % | [here] |
@misc{park2020aerial,
title={A Two-Stream Symmetric Network with Bidirectional Ensemble for Aerial Image Matching},
author={Jae-Hyun Park and Woo-Jeoung Nam and Seong-Whan Lee},
year={2020},
eprint={2002.01325},
archivePrefix={arXiv},
primaryClass={cs.CV}
}