https://github.com/seominseok0429/Self-Pair-for-Change-Detection

Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery (official)
https://github.com/seominseok0429/Self-Pair-for-Change-Detection

Last synced: 3 months ago
JSON representation

Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery (official)

• Host: GitHub
• URL: https://github.com/seominseok0429/Self-Pair-for-Change-Detection
• Owner: seominseok0429
• License: apache-2.0
• Created: 2022-12-20T12:52:10.000Z (almost 2 years ago)
• Default Branch: main
• Last Pushed: 2022-12-23T04:37:38.000Z (almost 2 years ago)
• Last Synced: 2024-07-23T02:41:54.397Z (4 months ago)
• Language: Python
• Size: 28.8 MB
• Stars: 25
• Watchers: 4
• Forks: 1
• Open Issues: 1
• Metadata Files:
  • Readme: README.md
  • License: LICENSE

Awesome Lists containing this project

• awesome-remote-sensing-change-detection - Seo M, Lee H, Jeon Y, et al. Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery

README

        
# Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery ([WACV2023](https://arxiv.org/abs/2212.10236))



  



## Update Note

- 2022.12.22 : Update [VSAIT](https://github.com/facebookresearch/vsait) style transfer method 

- This repository will be completed on the date of the WACV 2023 presentation.

## Install & Train (VSAIT-style transfer)

Sample



  



prepare

```

cd vsait

pip install -r requirements.txt

|./data/

|------/source

|-------------/train/A

|-------------/val/A

|------/target

|-------------/train/B

|-------------/val/B

python train.py --name="vsait"

python test.py --name="vsait_adapt" --checkpoint="./checkpoints/vsait/version_0/checkpoints/epoch={i}-step={j}.ckpt"

```

## Install (Open-CD)

```

pip install -U openmim

mim install mmcv-full

git clone https://github.com/seominseok0429/Self-Pair-for-Change-Detection.git

cd open-cd

pip install -v -e .

```

#### train

```

python tools/train.py configs/changer/changer_ex_r18_512x512_40k_levircd.py --work-dir ./changer_r18_levir_workdir --gpu-id 0 --seed 307

```

## Results

| Method    | Backbone | Dataset | Param (M)  | MACs (G)  | Precision |  Recall | F1 | config                                                                                                                                 | 

| --------- | -------- | --------- | ------: | -------: | -------------- | ----: | ------------- | -------------------------------------------------------------------------------------------------------------------------------------- | 

| Changer (ChangerEx) | ResNet18   | LEVIR-CD |  11.39 |     23.71 | 92.24           | 91.09 |  91.66        |  changer_ex_r18_512x512_40k_levircd.py

| Changer + Self-pair (ChangerEx) | ResNet18   | LEVIR-CD |  11.39 |     23.71 | TODO           | TODO |  TODO        |  TODO

## Citation

If you find this project useful in your research, please consider cite:

```bibtex

@article{seo2023selfpair,

  title={Self-Pair: Synthesizing Changes from Single Source for Object Change Detection in Remote Sensing Imagery}, 

  author={Minseok Seo, Hakjin Lee, Yongjin Jeon, Junghoon Seo},

  year={2023},

  eprint={2212.10236},

  archivePrefix={arXiv},

  primaryClass={cs.CV}

}

@article{fang2022changer,

  title={Changer: Feature Interaction is What You Need for Change Detection}, 

  author={Sheng Fang and Kaiyu Li and Zhe Li},

  year={2022},

  eprint={2209.08290},

  archivePrefix={arXiv},

  primaryClass={cs.CV}

}

@misc{opencd2022,

    title={{Open-CD}: An open source change detection toolbox},

    author={Open-CD Contributors},

    howpublished = {\url{https://github.com/likyoo/open-cd}},

    year={2022}

}

```

## Thank to

This code is heavily based on [open-cd](https://github.com/likyoo/open-cd).

We thank the authors of that code.