Shinoaki0145/SyncRepo

SynFoC

1. Introduction

This repository contains the implementation of the paper Steady Progress Beats Stagnation: Mutual Aid of Foundation and Conventional Models in Mixed Domain Semi-Supervised Medical Image Segmentation

In Conference on Computer Vision and Pattern Recognition (CVPR), 2025

2. Dataset Construction

The dataset needs to be divided into two folders for training and testing. The training and testing data should be in the format of the "data_format" folder.

3. Train

code/train.py is the implementation of our method .

Modify the paths in lines 770 to 817 of the code.

if args.dataset == 'fundus':
    train_data_path='../../data/Fundus' # the folder of fundus dataset
elif args.dataset == 'prostate':
    train_data_path="../../data/ProstateSlice" # the folder of prostate dataset
elif args.dataset == 'MNMS':
    train_data_path="../../data/mnms" # the folder of M&Ms dataset
elif args.dataset == 'BUSI':
    train_data_path="../../data/Dataset_BUSI_with_GT" # the folder of BUSI dataset

then simply run:

python train.py --dataset ... --lb_domain ... --lb_num ... --save_name ... --gpu 0 --AdamW --warmup --model MedSAM

4. Test

To run the evaluation code, please update the path of the dataset in test.py:

Modify the paths in lines 248 to 283 of the code.

then simply run:

python test.py --dataset ... --save_name ... --gpu 0

5. Summarize Logs

Use code/summarize_logs.py to summarize all log.txt files under model/ into one output file.

Run from the code/ directory:

python summarize_logs.py --model-dir ../model --format json --output summary.json

Other supported output formats:

python summarize_logs.py --model-dir ../model --format csv --output summary.csv
python summarize_logs.py --model-dir ../model --format markdown --output summary.md

The summary includes experiment metadata, last iteration/epoch, best UNet/SAM average Dice, final average Dice, last loss, and last mask ratio for each discovered log.

6. DataSets

Prostate Fundus M&Ms BUSI

The Prostate and M&Ms datasets have undergone preprocessing in our work, with the original data sourced from prostate and M&Ms

7. Acknowledgement

This project is based on the code from the SSL4MIS and SAMed project.

Thanks a lot for their great works.