AULUNet

This is our official code repository for "AULUNet: An Adaptive Ultra-Lightweight UNet Framework for Efficient Skin Lesion Segmentation in Resource-Constrained Environments".
Accepted to BMVC2025 as Oral (Selected only 3.4%) -- [Paper Link]

It covers environment setup, installing dependencies, preparing datasets, and training and testing.

1. Environment Setup & Install Dependencies

Create a Conda environment with Python 3.9:

conda create -n aulunet python=3.9 -y
conda activate aulunet

Upgrade pip (optional, but recommended):
```
pip install --upgrade pip
```
Install dependencies from requirements.txt:
```
pip install -r requirements.txt
```
If conflicts occur, recreate or adjust package versions.
Install PyTorch with CUDA 12.2:
```
pip install torch torchvision torchaudio --index-url https://download.pytorch.org/whl/cu121
```
This installs a PyTorch build (e.g., 2.5.1+cu121) compatible with CUDA 12.2.
Verify installation:
```
python -c "import torch; print(torch.__version__); print(torch.cuda.is_available())"
```
- Expected output includes a PyTorch version and True for CUDA availability.
- If False, confirm you have the correct NVIDIA drivers and CUDA.

2. Dataset Preparation

Depending on your dataset, you may need additional preprocessing. For certain legacy dependencies (e.g., scipy==1.2.1), consider using a separate Python 3.7 environment:

conda create -n tool python=3.7 -y
conda activate tool
pip install h5py
conda install scipy==1.2.1
pip install pillow

You can follow the UltraLight-VM-UNet for data preparation.

ISIC2017

Download from ISIC 2017 Challenge.
Extract training and ground-truth folders into /data/ISIC2017/.
Run:
```
python prepare_ISIC2017.py
```

ISIC2018

Download from ISIC 2018 Challenge.
Extract training and ground-truth folders into /data/ISIC2018/.
Run:
```
python prepare_ISIC2018.py
```

Custom Dataset

Organize files as:

./custom_data/
 ├── images/
 │   ├── xx0.png
 │   ├── xx1.png
 ├── masks/
 │   ├── xx0.png
 │   ├── xx1.png
 ├── prepare_custom_dataset.py

Adjust prepare_custom_dataset.py for train, validation, and test split sizes.
Run:
```
python prepare_custom_dataset.py
```

3. Training and Testing

3.1 Training AULUNet

Run the training script:

python train.py

By default, the training script may automatically run a test at the end of each epoch (or at final epochs). If you prefer to run tests separately or later, you can disable that section in train.py and use test.py whenever desired.
Outputs (logs, checkpoints, etc.) appear in ./results/.

If pretrained weights are needed, ensure you have them referenced in train.py or elsewhere.

3.2 Testing AULUNet

Update the checkpoint path in test.py (e.g., resume_model) to your trained model.
Run:
```
python test.py
```
Outputs will be saved in ./results/.

4. Acknowledgements

Thank you to the ISIC 2017/2018 Challenge organizers and UltraLight-VM-UNet
for sharing their works.

Enjoy experimenting with AULUNet! If you encounter any issues, consider opening an issue or discussion in the repository.

maklachur/AULUNet