LSST

This is the official PyTorch implementation of our paper:

Simple and Efficient: A Semisupervised Learning Framework for Remote Sensing Image Semantic Segmentation
Xiaoqiang Lu, Licheng Jiao, Fang Liu, Shuyuan Yang, Xu Liu, Zhixi Feng, Lingling Li, Puhua Chen
Accepted by IEEE Transactions on Geoscience and Remote Sensing (TGRS) 2022

Getting Started

Install

conda create -n lsst python=3.7
pip install -r requirements.txt

Data Preparation

Pre-trained Model

mkdir pretrained
cd pretrained
wget https://download.pytorch.org/models/resnet101-63fe2227.pth
mv resnet101-63fe2227.pth resnet101.pth
cd ..

or download via the following link ResNet-101

Dataset

We have processed the original dataset as mentioned in the paper. You can access the processed dataset directly via the following link.

GID-15 | iSAID | DFC22 | MER | MSL | Vaihingen

File Organization

├── ./pretrained
    └── resnet101.pth
    
├── [Your Dataset Path]
    ├── images
    └── labels

Results

Dataset	Setting	Method	mIoU
GID-15	1-8	baseline	61.86
		ours	66.38
	1-4	baseline	67.90
		ours	71.28
iSAID	100	baseline	39.91
		ours	46.94
	300	baseline	60.47
		ours	63.40
DFC22	1-8	baseline	26.97
		ours	30.94
	1-4	baseline	32.67
		ours	36.40
MER	1-8	baseline	43.63
		ours	49.68
	1-4	baseline	48.19
		ours	51.31
MSL	1-8	baseline	50.17
		ours	54.72
	1-4	baseline	50.26
		ours	56.22
Vaihingen	1-8	baseline	53.30
		ours	64.09
	1-4	baseline	59.30
		ours	65.34

Training and Testing

Training

Change DATASET, SPLIT, and DATASET_PATH as you want in train.py, then run:

CUDA_VISIBLE_DEVICES=0,1 python train.py

Testing

Change WEIGHTS, and DATASET_PATH as you want in test.py, then run:

CUDA_VISIBLE_DEVICES=0,1 python test.py

Acknowledgement

The code is mainly inherited from ST++, Thanks a lot for their great works!

Citation

If you find this project useful, please consider citing:

@article{lu2022simple,
  title={Simple and Efficient: A Semisupervised Learning Framework for Remote Sensing Image Semantic Segmentation},
  author={Lu, Xiaoqiang and Jiao, Licheng and Liu, Fang and Yang, Shuyuan and Liu, Xu and Feng, Zhixi and Li, Lingling and Chen, Puhua},
  journal={IEEE Transactions on Geoscience and Remote Sensing},
  volume={60},
  pages={1--16},
  year={2022},
  publisher={IEEE}
}