GitHub - zmoka-zht/FoBa

FoBa: A Foreground-Background co-Guided Method and New Benchmark for Remote Sensing Semantic Change Detection

Haotian Zhang¹, Han Guo¹, Keyan Chen¹, Hao Chen², Zhengxia Zou¹, Zhenwei Shi^{1, *}

¹ Beihang University, ² Shanghai Artificial Intelligence Laboratory.

^* Corresponding author

Overview | Get Started | Taken Away | Others | 简体中文版

🛎️Updates

Notice: FoBa has been accepted by IEEE TGRS! We'd appreciate it if you could give this repo a ⭐️star⭐️ and stay tuned!!
November 18th, 2025: FoBa has been accepted by IEEE TGRS!!

🔭Overview

FoBa serves as a strong benchmark for semantic change detection tasks.
The LevirSCD dataset

The architecture of FoBa

🗝️Let's Get Started!

`A. Installation`

Note that the code in this repo runs under Linux system. We have not tested whether it works under other OS.

The repo is based on the VMama repo and ChangeMamba repo, thus you need to install it first. The following installation sequence is taken from the VMamba repo.

Step 1: Clone the repository:

Clone this repository and navigate to the project directory:

git clone https://github.com/zmoka-zht/FoBa.git
cd FoBa

Step 2: Environment Setup:

It is recommended to set up a conda environment and installing dependencies via pip. Use the following commands to set up your environment:

Create and activate a new conda environment

conda create -n foba
conda activate foba

Install dependencies

pip install -r requirements.txt
cd kernels/selective_scan && pip install .

Dependencies for "Detection" and "Segmentation" (optional in VMamba)

pip install mmengine==0.10.1 mmcv==2.1.0 opencv-python-headless ftfy regex
pip install mmdet==3.3.0 mmsegmentation==1.2.2 mmpretrain==1.2.0

`B. Download Pretrained Weight`

Also, please download the pretrained weights of VMamba-Tiny, VMamba-Small, and VMamba-Base and put them under

project_path/FoBa/pretrained_weight/

`C. Data Preparation`

Semantic change detection

The SECOND dataset is used for semantic change detection experiments. Please download it and make it have the following folder/file structure. Note that the land-cover maps are RGB images in the original SECOND dataset for visualization, you need to transform them into single-channel. Also, the binary change maps should be generated by yourself and put them into folder.

Or you are welcome to directly download and use our preprocessed SECOND, LevirSCD, JL1 dataset from the following links:

Baidu Netdisk: Download link
Extraction code: foba

📋 Click to expand: LEVIR-SCD Class ID and RGB Visualization Mappings

Class to ID Mapping:

{
    "background": 0,
    "Sparse woodland": 1,
    "Low vegetation": 2,
    "Woodland": 3,
    "Playground": 4,
    "Low building": 5,
    "General building": 6,
    "Unpaved Road": 7,
    "Bare Land": 8,
    "Construction land": 9,
    "Parking Lot": 10,
    "Others": 11,
    "River": 12,
    "Impervious surfaces": 13,
    "Paved Road": 14,
    "High building": 15,
    "Water": 16
}

Note: You can find the implementation of the visualization code here.

${DATASET_ROOT} # Dataset root directory, for example: /home/username/data/SECOND
├── train
│   ├── im1
│   │   ├──00001.png
│   │   ├──00002.png
│   │   ├──00003.png
│   │   ...
│   │
│   ├── im2
│   │   ├──00001.png
│   │   ... 
│   │
│   ├── label   # Binary change map
│   │   ├──00001.png 
│   │   ... 
│   │
│   ├── label1   # Land-cover map of T1
│   │   ├──00001.png 
│   │   ...  
│   │
│   └── label2   # Land-cover map of T2
│       ├──00001.png 
│       ...  
│   
├── test
│   ├── ...
│   ...
├── list 
│   ├──train.txt
└── ├──test.txt

`D. Model Training`

Before training models, please enter into [changedetection] folder, which contains all the code for network definitions, training and testing.

cd <project_path>/FoBa/changedetection

Semantic change detection

The following commands show how to train and evaluate FoBa on the SECOND dataset:

python script/train_foba.py  --dataset 'SECOND' \
                                 --batch_size 2 \
                                 --crop_size 512 \
                                 --max_iters 480000 \
                                 --model_type FoBa \
                                 --model_param_path '<project_path>/FoBa/changedetection/saved_models' \ 
                                 --train_dataset_path '<dataset_path>/SECOND/train' \
                                 --train_data_list_path '<dataset_path>/SECOND/list/train_list.txt' \
                                 --test_dataset_path '<dataset_path>/SECOND/test' \
                                 --test_data_list_path '<dataset_path>/SECOND/list/test_list.txt'
                                 --cfg '<project_path>/FoBa/changedetection/configs/vssm1/vssm_small_224.yaml' \
                                 --pretrained_weight_path '<project_path>/FoBa/pretrained_weight/vssm_small_0229_ckpt_epoch_222.pth'

`E. Inference Using Our/Your Weights`

Before inference, please enter into [changedetection] folder.

cd <project_path>/FoBa/changedetection

Semantic change detection

The following commands show how to infer semantic change maps using trained FoBa on the SECOND dataset:

python script/infer_foba_second.py  --dataset 'SECOND'  \
                                 --model_type 'FoBaMambaBased' \
                                 --test_dataset_path '<dataset_path>/SECOND/test' \
                                 --test_data_list_path '<dataset_path>/SECOND/list\test_list.txt' \
                                 --cfg '<project_path>/FoBa/changedetection/configs/vssm1/vssm_base_224.yaml' \
                                 --pretrained_weight_path '<project_path>/FoBa/pretrained_weight/vssm_base_0229_ckpt_epoch_237.pth'
                                 --resume '<saved_model_path>/[your_trained_model].pth'

⚗️Results Taken Away

We'd appreciate it if you could give this repo a ⭐️star⭐️ and stay tuned.

`A. Pretrained Weight of VMamba (Encoder)`

Method	ImageNet (ckpt)
VMamba-Tiny	[GDrive]
VMamba-Small	[GDrive]
VMamba-Base	[GDrive]

`B. Semantic Change Detection`

Method	Checkpoints
FoBaMambaBased	SECOND + LevirSCD + JL1 (Extraction code: `foba`)
FoBaTransformerBased	SECOND + LevirSCD + JL1 (Extraction code: `foba`)

📜Reference

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

@ARTICLE{11268372,
  author={Zhang, Haotian and Guo, Han and Chen, Keyan and Chen, Hao and Zou, Zhengxia and Shi, Zhenwei},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={FoBa: A Foreground-Background co-Guided Method and New Benchmark for Remote Sensing Semantic Change Detection}, 
  year={2025},
  volume={},
  number={},
  pages={1-1},
  keywords={Semantics;Remote sensing;Transformers;Feature extraction;Annotations;Roads;Multitasking;Spatial resolution;Landsat;Land surface;Semantic change detection (SCD);foreground-background co-guided;bi-temporal interaction;mamba;new benchmark},
  doi={10.1109/TGRS.2025.3636947}}

@ARTICLE{10902569,
  author={Zhang, Haotian and Chen, Keyan and Liu, Chenyang and Chen, Hao and Zou, Zhengxia and Shi, Zhenwei},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={CDMamba: Incorporating Local Clues Into Mamba for Remote Sensing Image Binary Change Detection}, 
  year={2025},
  volume={63},
  number={},
  pages={1-16},
  keywords={Feature extraction;Transformers;Remote sensing;Convolutional neural networks;Visualization;Artificial intelligence;Spatiotemporal phenomena;Computational modeling;Attention mechanisms;Computer vision;Bi-temporal interaction;change detection (CD);high-resolution optical remote sensing image;Mamba;state-space model},
  doi={10.1109/TGRS.2025.3545012}}

@ARTICLE{10471555,
  author={Zhang, Haotian and Chen, Hao and Zhou, Chenyao and Chen, Keyan and Liu, Chenyang and Zou, Zhengxia and Shi, Zhenwei},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={BiFA: Remote Sensing Image Change Detection With Bitemporal Feature Alignment}, 
  year={2024},
  volume={62},
  number={},
  pages={1-17},
  keywords={Feature extraction;Task analysis;Remote sensing;Transformers;Interference;Decoding;Optical flow;Bitemporal interaction (BI);change detection (CD);feature alignment;flow field;high-resolution optical remote sensing image;implicit neural representation},
  doi={10.1109/TGRS.2024.3376673}}

🤝Acknowledgments

This project is based on VMamba (paper, code), ChangeMamba (paper, code). Thanks for their excellent works!!

🙋Q & A

For any questions, please feel free to contact us.

Name		Name	Last commit message	Last commit date
Latest commit History 35 Commits
.idea		.idea
analyze		analyze
changedetection		changedetection
classification		classification
figures		figures
kernels/selective_scan		kernels/selective_scan
.gitignore		.gitignore
LICENSE		LICENSE
README.md		README.md
README_zh-CN.md		README_zh-CN.md
modelcard.sh		modelcard.sh
requirements.txt		requirements.txt

Navigation Menu

Repository files navigation

FoBa: A Foreground-Background co-Guided Method and New Benchmark for Remote Sensing Semantic Change Detection

🛎️Updates

🔭Overview

🗝️Let's Get Started!

`A. Installation`

`B. Download Pretrained Weight`

`C. Data Preparation`

`D. Model Training`

`E. Inference Using Our/Your Weights`

⚗️Results Taken Away

`A. Pretrained Weight of VMamba (Encoder)`

`B. Semantic Change Detection`

📜Reference

🤝Acknowledgments

🙋Q & A

Star History

About

Releases

Packages

Contributors

Languages

Folders and files

Latest commit

History

Repository files navigation

FoBa: A Foreground-Background co-Guided Method and New Benchmark for Remote Sensing Semantic Change Detection

🛎️Updates

🔭Overview

🗝️Let's Get Started!

A. Installation

B. Download Pretrained Weight

C. Data Preparation

D. Model Training

E. Inference Using Our/Your Weights

⚗️Results Taken Away

A. Pretrained Weight of VMamba (Encoder)

B. Semantic Change Detection

📜Reference

🤝Acknowledgments

🙋Q & A

Star History

About

Resources

Stars

Watchers

Forks

Releases

Packages

Contributors

Languages

`A. Installation`

`B. Download Pretrained Weight`

`C. Data Preparation`

`D. Model Training`

`E. Inference Using Our/Your Weights`

`A. Pretrained Weight of VMamba (Encoder)`

`B. Semantic Change Detection`