Full-scale Representation Guided Network for Retinal Vessel Segmentation

Official repository of the paper Full-scale Representation Guided Network for Retinal Vessel Segmentation

Environment

• OS: Ubuntu 22.04 LTS
• GPU: RTX 4090 24GB
• GPU Driver version: 550.54.14
• CUDA: 12.4
• Pytorch 2.4.1

✅ Experimental Result

Dataset	mIoU	F1 score	Acc	AUC	Sen	MCC
DRIVE	84.068	83.229	97.042	98.235	84.207	81.731
STARE	86.118	85.100	97.746	98.967	86.608	83.958
CHASE_DB1	82.680	81.019	97.515	99.378	85.995	79.889
HRF	83.088	81.567	97.106	98.744	83.616	80.121

✅ Pretrained model for each dataset

Each pre-trained model could be found on release version

🧻 Dataset Preparation

You can edit train_x_path... in configs/train.yml
The input and label should be sorted by name, or the dataset is unmatched to learn.

For train/validation set, you can download from public link or release version

---

🚄 Train

If you have installed 'WandB', login your ID in command line.
If not, modify wandb=false in configs/train.yml.
You can login through your command line or wandb.login() inside "main.py"

For Train, edit the configs/train.yml and execute below command

bash bash_train.sh

---

🛴 Inference

For Inference, edit the configs/inference.yml and execute below command.
Please locate your model path via model_path in configs/inference.yml

bash bash_inference.sh

• If you are using pretrained model, the result should be approximate to experimental result's

📚 Citation

@article{seo2025fullscalerepresentationguidednetwork,
  title   = {Full-scale Representation Guided Network for Retinal Vessel Segmentation},
  author  = {Sunyong Seo, Huisu Yoon, Semin Kim, Jongha Lee},
  journal = {arXiv preprint arXiv:2501.18921},
  year    = {2025}
}