[CVPR 2024] Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining

The official implementation of "Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining".

More detailed information is in the PAPER.

Authors: Jiahao Nie^* Yun Xing^*, Gongjie Zhang, Pei Yan, Aoran Xiao, Yap-Peng Tan, Alex C. Kot, Shijian Lu

Datasets

The following datasets are used for evaluation in CD-FSS:

Source domain:

• 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].

Target domains:

• Deepglobe:

  Home: http://deepglobe.org/

  Direct: https://www.kaggle.com/datasets/balraj98/deepglobe-land-cover-classification-dataset

  Data Preprocessing Code: Please refer preprocess_deepglobe.py or PATNet repo.

• ISIC2018:

  Home: http://challenge2018.isic-archive.com

  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:

  Home: https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4256233/

  Direct: https://www.kaggle.com/datasets/nikhilpandey360/chest-xray-masks-and-labels

• FSS-1000:

  Home: https://github.com/HKUSTCV/FSS-1000

  Direct: https://drive.google.com/file/d/16TgqOeI_0P41Eh3jWQlxlRXG9KIqtMgI/view

Pre-trained ResNet Models

Download pre-trained ResNet models: GoogleDrive

Performance and Trained Models

Download trained models: GoogleDrive

Datasets	Setting	mIOU	Setting	mIOU
Deepglobe	1-Shot	50.6%	5-Shot	58.8%
ISIC	1-Shot	66.3%	5-Shot	69.8%
Chest X-Ray	1-Shot	74.0%	5-Shot	74.6%
FSS-1000	1-Shot	80.1%	5-Shot	82.4%

File Organization

    IFA/                                                # project file
    ├── dataset/                                        # dataset
    |   ├── VOC2012/                                    # source dataset: pascal voc 2012
    |   |   ├── JPEGImages/
    |   |   └── SegmentationClassAug/
    |   ├── Deeoglobe                                   # target dataset: deepglobe
    |   |   ├── 01_train_ori/                           # original data
    |   |   ├── ...
    |   |   └── 04_train_cat/                           # processed data
    |   |       ├── 1/                                  # category
    |   |       |   └── test/
    |   |       |       ├── origin/                     # image
    |   |       |       └── groundtruth/                # mask
    |   |       └── ...    
    |   ├── 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
    |   └── resnet101.pth
    |
    └── trained_models/                                 # official trained models
        ├── deepglobe/                                  # target dataset
        └── ...

Run the code

Please refer the scripts in all.sh to train, fine-tune, and evaluate your models.

Here is an example on Deepglobe dataset: First, you need to train a model on the source dataset:

CUDA_VISIBLE_DEVICES=0 python -W ignore train.py \
  --dataset deepglobe --data-root ./dataset \
  --backbone resnet50 --batch-size 12 --shot 1 --refine

Then, you need to use IFA to fine-tuning the trained model on the target dataset:

CUDA_VISIBLE_DEVICES=0 python -W ignore ifa.py \
  --dataset deepglobe --data-root ./dataset \
  --backbone resnet50 --batch-size 12 --shot 1 --refine --lr 0.0005

You can use our trained models for evaluation directly:

CUDA_VISIBLE_DEVICES=0 python -W ignore test.py \
  --dataset deepglobe --data-root ./dataset \
  --backbone resnet50 --batch-size 96 --shot 1 --refine

Please note that the performances may flutuate within a small range because of differnet batch-sizes, seeds, devices, and environments.

Citation

If you use this codebase for your research, please consider citing:

@article{nie2024cross,
  title={Cross-Domain Few-Shot Segmentation via Iterative Support-Query Correspondence Mining},
  author={Nie, Jiahao and Xing, Yun and Zhang, Gongjie and Yan, Pei and Xiao, Aoran and Tan, Yap-Peng and Kot, Alex C and Lu, Shijian},
  journal={arXiv preprint arXiv:2401.08407},
  year={2024}
}

Acknowledgement

Our codebase is built based on PATNet and SSP's official code.

We also thank HSNet and other FSS and CD-FSS works for their great contributions.

Reference

[1] Shuo Lei, Xuchao Zhang, Jianfeng He, Fanglan Chen, Bowen Du, and Chang-Tien Lu. Cross-domain few-shot semantic segmentation. ECCV, 2022.

[2] Qi Fan, Wenjie Pei, Yu-Wing Tai, and Chi-Keung Tang. Self-support few-shot semantic segmentation. ECCV, 2022.

[3] Juhong Min, Dahyun Kang, and Minsu Cho. Hypercorrelation squeeze for few-shot segmentation. ICCV, 2021.