Prior-Guided Selective Parameter Fine-Tuning for Source-Free Domain Adaptive Medical Image Segmentation

PyTorch implementation of our source-free unsupervised domain adaptation method with prior-guided selective parameter fine-tuning.

Data Preparation

Dataset Structure

Ensure your data directory follows this structure:

data_root/
├── Domain1/
│   ├── train/ROIs/
│   └── test/ROIs/
├── Domain2/
│   ├── train/ROIs/
│   └── test/ROIs/
└── Domain3/
    ├── train/ROIs/
    └── test/ROIs/

Training

1. Source Domain Training

First, train the model on the source domain:

python train_source.py \
    --gpu 0 \
    --dataset Domain3 \
    --data-dir /path/to/your/data \
    --batch-size 8 \
    --max-epoch 200

2. Target Domain Adaptation

Then, perform source-free domain adaptation on the target domain:

python train_target.py \
    --gpu 0 \
    --model-file /path/to/source/model.pth.tar \
    --out /path/to/output/directory \
    --data-dir /path/to/your/data \
    --dataset Domain2 \
    --batch-size 8 \
    --epoch 30 \
    --keep-ratio 0.1 \
    --import-ratio 0.1