MemoryAdaptNet-master:Unsupervised Domain Adaptation Semantic Segmentation of High Resolution Remote Sensing Imagery with Invariant Domain-level Prototype Memory

Pytorch implementation of our method for cross-domain semantic segmentation of the high-resolution remote sensing imagery.

Contact: Jingru Zhu (zhujingru@csu.edu.cn)

Paper

Unsupervised Domain Adaptation Semantic Segmentation of HRS Imagery with Invariant Domain-level Prototype Memory
Jingru Zhu, Ya Guo , Geng Sun, Lobo Yang, Min Deng and Jie Chen, Member, IEEE,
IEEE Transactions on Geoscience and Remote Sensing, 2023.

Example Results

Quantitative Reuslts

Installation

• Install PyTorch from http://pytorch.org with Python 3.6 and PyTorch 1.8.0

• Clone this repo

  git clone https://github.com/RS-CSU/MemoryAdaptNet-master
  cd MemoryAdaptNet-master

  Dataset

• Download the Potsdam Dataset as the source domain, and put it in the dataset/Potsdam folder

• Download the Vaihingen Dataset as the target domain, and put it in the dataset/Vaihingen folder

Testing

• Download the pre-trained p2v_best_m.pth and put it in the snapshots folder

• Download the pre-trained p2v_best.pth and put it in the snapshots folder

• Test the model and results will be saved in the results folder

• The ablation models

• Download the pre-trained p2v_IDMA_m.pth and put it in the snapshots folder

• Download the pre-trained p2v_IDMA.pth and put it in the snapshots folder

• Download the pre-trained p2v_FDA+IDMA_m.pth and put it in the snapshots folder

• Download the pre-trained p2v_FDA+IDMA.pth and put it in the snapshots folder

  python test_p2v_v3_1.py

Training Examples

• Train the Potsdam(IR-R-G) to Vaihingen(IR-R-G) model

python train_p2v_v3_1.py

Related Implementation and Dataset

Acknowledgment

This code is heavily borrowed from Pytorch-AdaptSegNet and DeepLabv3Plus-Pytorch.

Note

The model and code are available for non-commercial research purposes only.

• 07/2022: code released