Home

Awesome

GeleNet

This project provides the code and results for 'Salient Object Detection in Optical Remote Sensing Images Driven by Transformer', IEEE TIP, 2023. IEEE and arxiv Homepage

Network Architecture

<div align=center> <img src="https://github.com/MathLee/GeleNet/blob/main/images/GeleNet.png"> </div>

Requirements

python 3.8 + pytorch 1.9.0

Saliency maps

We provide saliency maps of our GeleNet on three datasets in './GeleNet_saliencymap_PVT.zip' (PVT-v2-b2 backbone) and './GeleNet_saliencymap_SwinT.zip' (Swin Transformer backbone).

We also provide saliency maps of all compared methods (code: 2892) on three datasets.

Image

Training

We use data_aug.m for data augmentation.

Download pvt_v2_b2.pth (code: sxiq), and put it in './model/'.

Modify paths of datasets, then run train_GeleNet.py.

Note: Our main model is under './model/GeleNet_models.py' (PVT-v2-b2 backbone)

Pre-trained model and testing

  1. Download the pre-trained models (PVT-v2-b2 backbone) on ORSSD (code: qga2), EORSSD (code: ahm7), and ORSI-4199 (code: 5h3u), and put them in './models/'.

  2. Modify paths of pre-trained models and datasets.

  3. Run test_GeleNet.py.

Evaluation Tool

You can use the evaluation tool (MATLAB version) to evaluate the above saliency maps.

ORSI-SOD_Summary

Citation

    @ARTICLE{Li_2023_GeleNet,
            author = {Gongyang Li and Zhen Bai and Zhi Liu and Xinpeng Zhang and Haibin Ling},
            title = {Salient Object Detection in Optical Remote Sensing Images Driven by Transformer},
            journal = {IEEE Transactions on Image Processing},
            volume = {32},
            pages = {5257-5269},
            year = {2023},
            }
            
            

If you encounter any problems with the code, want to report bugs, etc.

Please contact me at lllmiemie@163.com or ligongyang@shu.edu.cn.