Skip to content

[ISPRS J P&RS 2023] Exchange means change: an unsupervised change detection framework based on intra- and inter-image patch exchange

Notifications You must be signed in to change notification settings

ChenHongruixuan/I3PE

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

8 Commits
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

Exchange means change: an unsupervised single-temporal change detection framework based on intra- and inter-image patch exchange

This is an official implementation of I3PE framework in our ISPRS JP&RS 2023 paper: Exchange means change: an unsupervised single-temporal change detection framework based on intra- and inter-image patch exchange.



Get started

Requirements

Please download the following key python packages in advance.

python==3.6.15
pytorch==1.7.0
scikit-learn==0.22.1
scikit-image==0.17.2
imageio=2.15.0
numpy==1.19.5
tqdm==4.64.1

Datasets

Two large-scale benchmark datasets, SYSU dataset and SECOND dataset, are used for experiments. Please download them and organize them in the following way.

For the Wuhan dataset used in our paper, you can also download it here for your own research [Google Drive], [Baidu Cloud].

├── <THE-ROOT-PATH-OF-DATA>/
│   ├── SYSU/     
|   |   ├── train/
|   |   |   ├── T1/
|   |   |   ├── T2/
|   |   |   ├── GT/
|   |   ├── val/
|   |   |   |── ...
|   |   ├── test/
|   |   |   |── ...
|   |   
│   ├── SECOND/     
|   |   ├── train/
|   |   |   |── ...
|   |   ├── test/
|   |   |   |── ...

Generate Single-Temporal Training Sets

Transfer the images in T1 and T2 under original training set to a new folder and rename the images.

python construct_single_temporal_set.py

Generate object maps and cluster maps for intra-image patch exchange method in advance.

python generate_object.py --dataset_path 'your own path here' --obj_num 1000
python generate_clustering_map.py --dataset_path 'your own path here' --eps 7 --min_samples 10

Training Change Detectors

Training the deep change detector on the single-temporal training sets using both intra- and inter-image patch exchangem methods.

python train_network_I3PE.py

Unsupervised change detection results of different methods on the test sets:

SYSU SECOND
Method OA F1 OA F1
CVA 0.4539 0.3492 0.4332 0.3003
IRMAD 0.6914 0.3705 0.6829 0.3451
ISFA 0.6977 0.3695 0.7130 0.3293
OBCD 0.7091 0.4046 0.7005 0.3426
DCAE 0.7636 0.4390 0.7600 0.3340
DCVA 0.6995 0.4450 0.6795 0.3681
DSFA 0.6326 0.4125 0.5961 0.3301
KPCA-MNet 0.7084 0.4482 0.6793 0.3670
I3PE 0.7305 0.5547 0.7283 0.4380

Citation

If this code or dataset contributes to your research, please kindly consider citing our paper :)

@article{Chen2023Exchange,
    title = {Exchange means change: An unsupervised single-temporal change detection framework based on intra- and inter-image patch exchange},
    author = {Hongruixuan Chen and Jian Song and Chen Wu and Bo Du and Naoto Yokoya},
    journal = {ISPRS Journal of Photogrammetry and Remote Sensing},
    volume = {206},
    pages = {87-105},
    year = {2023},
    issn = {0924-2716},
    doi = {https://doi.org/10.1016/j.isprsjprs.2023.11.004}
}

Q & A

For any questions, please contact us.

Releases

No releases published

Packages

No packages published

Languages