Source code for "**SP-RAN: Self-Paced Residual Aggregated Network for Solar Panel Mapping in Weakly Labeled Aerial Images **", accepted in TGRS. The paper's PDF can be found in Here.
Jue Zhang; Xiuping Jia; Jiankun Hu
School of Engineering and Information Technology, University of New South Wales, Canberra
- Windows 10
- Torch 1.12.0
- CUDA 11.6.0
- Python 3.7.13
- Opencv 3.4.2
GoogleEarth Static Map API
Please refer to act_config.json
GradCAM
Setting the training data to the proper root as follows:
-- data -- train -- fore
-- back
-- Pseudo labels
-- test -- cls -- fore
-- back
-- seg -- img
-- gt
set self-pace
to False
in act_config.json
.
Run python train.py --config_path act_config.json
.
Set self-pace
to True
in act_config.json
. Set the label update dir in act_config.json
. Run python train.py --config_path act_config.json
.
python predict.py --config_path act_config.json
Noting that the results in our paper do not adopt any post-process including CRF.
The evaluation code can be found in here.
If you have any questions, pleas feel free to contact me: jue.zhang@adfa.edu.au.
We really hope this repo can contribute the conmunity, and if you find this work useful, please use the following citation:
@ARTICLE{9585690,
author={Zhang, Jue and Jia, Xiuping and Hu, Jiankun},
journal={IEEE Transactions on Geoscience and Remote Sensing},
title={SP-RAN: Self-Paced Residual Aggregated Network for Solar Panel Mapping in Weakly Labeled Aerial Images},
year={2022},
volume={60},
number={},
pages={1-15},
doi={10.1109/TGRS.2021.3123268}}