CDKD--Knowledge Distillation-Based Lightweight Change Detection in High-Resolution Remote Sensing Imagery for On-Board Processing

The pytorch codes for IEEE JSTARS paper "Knowledge Distillation-Based Lightweight Change Detection in High-Resolution Remote Sensing Imagery for On-Board Processing".

Requirements

• Python 3.8
• Pytorch 1.9

Dataset format

• CDD/SYSU-CD/WHU-CD
  • train
    • A
    • B
    • OUT
  • val
    • A
    • B
    • OUT
  • test
    • A
    • B
    • OUT

Train from scratch

python train.py

Evaluate model performance

python eval.py

Pre-trained teacher models

The pre-trained teacher models are available.

baidu disk (cdkd)

Citation

If you find this work valuable or use our code in your own research, please consider citing us with the following bibtex:

@ARTICLE{
  author={G. Wang, N. Zhang, J. Wang, W. Liu, Y. Xie and H. Chen},
  journal={IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing}, 
  title={Knowledge Distillation-Based Lightweight Change Detection in High-Resolution Remote Sensing Imagery for On-Board Processing}, 
  year={2024},
  volume={},
  number={},
  pages={}}