HappyKoala 🐨 & OTTER

A unified segmentation benchmark integrating advanced models on ISIC 2018 dataset.

Description

HappyKoala is an experimental framework designed for comprehensive evaluation of medical image segmentation models on the ISIC 2018 skin lesion segmentation dataset.
This project integrates multiple state-of-the-art architectures including classical UNet variants, attention-enhanced models, and modern Transformer-based designs.

By providing a unified training and evaluation pipeline, HappyKoala facilitates fair comparison, reproducibility, and rapid prototyping of segmentation models.

OTTER Publication Note

The OTTER submission has been accepted. The related code is located in the ddpm/ directory. The segmentation part uses the code from this HappyKoala project. The remaining parts are still being organized...

Haocheng Kan, Yuesheng Zhu, Guibo Luo, and Hanwen Zhang. OTTER: Optimized Training with Trustworthy Enhanced Replication via Diffusion and Federated VMUNet for Privacy-Aware Medical Segmentation [C]// Proceedings of the 27th International Conference on Information and Communications Security (ICICS). Nanjing, China: Springer Nature, October 2025. (CCF-C) DOI: 10.1007/978-981-95-3543-9_18

@inproceedings{kan2025otter,
  title={OTTER: Optimized Training with Trustworthy Enhanced Replication via Diffusion and Federated VMUNet for Privacy-Aware Medical Segmentation},
  author={Kan, Haocheng and Zhu, Yuesheng and Luo, Guibo and Zhang, Hanwen},
  booktitle={Information and Communications Security: 27th International Conference, ICICS 2025, Nanjing, China, October 29--31, 2025, Proceedings, Part II},
  pages={331--346},
  year={2025},
  organization={Springer},
  doi={10.1007/978-981-95-3543-9_18}
}

Links:

• Paper: https://dl.acm.org/doi/10.1007/978-981-95-3543-9_18
• ICICS 2025 Accepted Papers: https://www.icics2025.org/papers.html

Integrated Models

• UNet
• VMUNet
• U2Net
• UNet++
• UNet+++
• VMUNetV2
• HVMUNet
• TransUNet
• ResUNet
• ResUNet++

Features

• 📚 Unified training and evaluation pipeline
• 🏥 Focused on medical image segmentation (skin lesion)
• 🧩 Modular architecture: easily add new models
• 📊 Standard metrics (IoU, DSC, Accuracy, Sensitivity, Specificity)
• 📈 Visualization of segmentation results

Installation

# Clone the repository
git clone https://github.com/kancheng/happykoala.git

# Navigate to the project directory
cd happykoala

Environments

conda create -n vmunet python=3.8
conda activate vmunet
pip install torch==1.13.0 torchvision==0.14.0 torchaudio==0.13.0 --extra-index-url https://download.pytorch.org/whl/cu117
pip install packaging
pip install timm==0.4.12
pip install pytest chardet yacs termcolor
pip install submitit tensorboardX
pip install triton==2.0.0
pip install causal_conv1d==1.0.0  # causal_conv1d-1.0.0+cu118torch1.13cxx11abiFALSE-cp38-cp38-linux_x86_64.whl
pip install mamba_ssm==1.0.1  # mmamba_ssm-1.0.1+cu118torch1.13cxx11abiFALSE-cp38-cp38-linux_x86_64.whl
pip install scikit-learn matplotlib thop h5py SimpleITK scikit-image medpy yacs

The .whl files of causal_conv1d and mamba_ssm could be found here. {Baidu or GoogleDrive}

Dataset

• ISIC 2018
  • ISIC 2018: Skin Lesion Analysis Towards Melanoma Detection
  • https://challenge.isic-archive.com/data

ISIC datasets

• The ISIC17 and ISIC18 datasets, divided into a 7:3 ratio, can be found here {Baidu or GoogleDrive}.

• After downloading the datasets, you are supposed to put them into './external/isic2017/' and './external/isic2018/', and the file format reference is as follows. (take the ISIC17 dataset as an example.)

• './data/isic17/'

  • train
    • images
      • .png
    • masks
      • .png
  • val
    • images
      • .png
    • masks
      • .png

Demo: Model Outputs on ISIC 2018

Below are example segmentation results generated by each integrated model on the ISIC 2018 dataset. These visualizations help illustrate the qualitative differences in segmentation performance and boundary refinement across various architectures.

Figure 1. UNet

Figure 2. VMUNet

Figure 3. U2Net

Figure 4. UNet++

Figure 5. UNet+++

Figure 6. VMUNetV2

Figure 7. HVMUNet

Figure 8. TransUNet

Figure 9. ResUNet

Figure 10. ResUNet++

Acknowledgments

We would like to express our gratitude to the following open-source projects and authors whose work greatly contributed to this project:

• VM-UNet
  Provided the implementation and key insights for integrating Vision Mamba-based architectures into our segmentation models.

• U2Net
  Offered a well-documented reference implementation of deep nested U-shaped networks with residual U-blocks.

• TransUNet
  Contributed Transformer-based UNet designs which inspired the global-local feature modeling in our framework.

• Other original UNet-based and Transformer-based works cited in this project.

We sincerely thank all researchers and developers in the open-source community whose work enables rapid progress in medical image analysis.