🩺 FSG-Net: Full-scale Representation Guided Network for Retinal Vessel Segmentation

Official PyTorch Implementation

---

📋 Table of Contents

• Environment Setup
• Performance Results
• Pretrained Models
• Dataset Preparation
• Training
• Inference
• Citation

---

🖥️ Environment Setup

Component	Version
OS	Ubuntu 22.04 LTS
GPU	RTX 4090 (24GB)
GPU Driver	550.54.14
CUDA	12.4
PyTorch	2.4.1

---

🎯 Performance Results

Comprehensive evaluation across multiple retinal vessel segmentation datasets:

Dataset	mIoU	F1 Score	Accuracy	AUC	Sensitivity	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 Models

Pre-trained models for each dataset are available in the releases page.

Download and extract the models corresponding to your target dataset before inference.

---

📊 Dataset Preparation

Configuration

Edit the dataset paths in configs/train.yml:

• Update train_x_path with your input data directory
• Update corresponding label paths

Important Requirements

• ✋ Input and label files must be sorted by name
• ✋ Ensure one-to-one correspondence between inputs and labels

Download Datasets

Datasets can be obtained from:

• Public repositories
• GitHub Releases

---

🚀 Training

Step 1: Configure Training

Edit configs/train.yml with your:

• Dataset paths
• Hyperparameters
• Training settings

Step 2: Setup Monitoring (Optional)

With WandB:

wandb login  # Login with your credentials

Without WandB: Set wandb: false in configs/train.yml

Step 3: Start Training

bash bash_train.sh

---

🔍 Inference

Step 1: Configure Inference

Edit configs/inference.yml:

• Set model_path to your pretrained model location
• Specify input image directory
• Configure output paths

Step 2: Run Inference

bash bash_inference.sh

Expected Results

When using official pretrained models, results should closely match the performance metrics above.

---

📖 Citation

If you find this work helpful, please cite:

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

---

⭐ If you find this repository useful, please consider giving it a star!