The official implementation of "Divide-and-Conquer Decoupled Network for Cross-Domain Few-Shot Segmentation".
More detailed information is in the PAPER.
Authors: Runmin Cong, Anpeng Wang, Bin Wan, Cong Zhang, Xiaofei Zhou, Wei Zhang
The following datasets are used for evaluation in CD-FSS:
-
PASCAL VOC2012:
Download PASCAL VOC2012 devkit (train/val data):
wget http://host.robots.ox.ac.uk/pascal/VOC/voc2012/VOCtrainval_11-May-2012.tar
Download PASCAL VOC2012 SDS extended mask annotations from [Google Drive].
-
Deepglobe:
Data Preprocessing Code: Please refer preprocess_deepglobe.py or PATNet repo.
-
ISIC2018:
Direct (must login): https://challenge.isic-archive.com/data#2018
Class Information: data/isic/class_id.csv
Data Preprocessing Code: Please refer preprocess_isic.py or PATNet repo.
-
Chest X-ray:
-
FSS-1000:
Home: https://github.com/HKUSTCV/FSS-1000
For convenience, you can also download the target domain datasets directly from our organized Baidu Netdisk.
Download pre-trained ResNet models: GoogleDrive or Baidu Netdisk.
Download SSP pre-trained models: GoogleDrive or Baidu Netdisk.
DCDNet/ # project file
├── dataset/ # dataset
| ├── VOC2012/ # source dataset: pascal voc 2012
| | ├── JPEGImages/
| | └── SegmentationClassAug/
| ├── Deepglobe # target dataset: deepglobe
| | ├── 1/ # category
| | | └── test/
| | | ├── origin/ # image
| | | └── groundtruth/ # mask
| | ├── 2/ # category
| | └── ...
| ├── ISIC # target dataset: isic
| | ├── ISIC2018_Task1-2_Training_Input/ # image
| | | ├── 1/ # category
| | | └── ...
| | └── ISIC2018_Task1_Training_GroundTruth/ # mask
| | └── ...
| ├── LungSegmentation/ # target dataset: chest x-ray
| | ├── CXR_png/ # image
| | └── masks/ # mask
| └── FSS-1000 # target dataset: fss-1000
| ├── ab_wheel/ # category
| └── ...
|
├── pretrained/ # pretrained resnet models
| ├── resnet50.pth
| └── Ori_SSP_trained_on_VOC.pth
|
└── trained_models/ # official trained models
├── deepglobe/ # target dataset
└── ...
Conda environment settings:
conda create -n dcdnet python=3.10
conda activate dcdnet
pip install uv
# Install PyTorch 2.2.0 with CUDA 12.1
uv pip install torch==2.2.0 torchvision==0.17.0 torchaudio --index-url https://download.pytorch.org/whl/cu121
uv pip install -r requirements.txtHere is an example on ISIC dataset:
First, you need to train a model on the source dataset:
python train.py --data-root ./dataset --dataset isic --cuda 0
Then, you need to fine-tuning the trained model on the target dataset:
python finetuning.py --data-root ./dataset --dataset isic --cuda 0
You can use our trained models for evaluation directly:
python test.py --data-root ./dataset --dataset isic --cuda 0
Please note that the performances may flutuate within a small range because of differnet batch-sizes, seeds, devices, and environments.
If you use this codebase for your research, please consider citing:
@article{cong2025divide,
title={Divide-and-Conquer Decoupled Network for Cross-Domain Few-Shot Segmentation},
author={Cong, Runmin and Wang, Anpeng and Wan, Bin and Zhang, Cong and Zhou, Xiaofei and Zhang, Wei},
journal={arXiv preprint arXiv:2511.07798},
year={2025}
}
Our codebase is built based on IFA and SSP's official code.
We also thank PATNet and other FSS and CD-FSS works for their great contributions.
[1] Shuo Lei, Xuchao Zhang, Jianfeng He, Fanglan Chen, Bowen Du, and Chang-Tien Lu. Cross-domain few-shot semantic segmentation. ECCV, 2022.
[2] Jiahao Nie, Yun Xing, Gongjie Zhang, Pei Yan, Aoran Xiao, Yap-Peng Tan, Alex C Kot, Shijian Lu. Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining. CVPR, 2024.