DREAM-RFI Dataset v1.0.0 - Initial Publication Release

This release accompanies the paper "A Harmonized Multi-Source Dataset with Baseline Deep Learning Validation for Staging Diabetic Retinopathy" submitted to IJMEMS.

Dataset Overview

• Total Images: 3,461 quality-filtered retinal fundus images
• Classes: 5-class DR severity (No DR: 1256, Mild NPDR: 493, Moderate NPDR: 779, Severe NPDR: 396, PDR: 537)
• Source Datasets: IDRiD, Messidor-2, SUSTech-SYSU, APTOS, DeepDRiD-v1.1, Zenodo DR V03
• Standard Resolution: 1024×1024×3 pixels
• Grading Standard: ICDR severity scale (0-4)

Repository Contents

• Complete quality assessment pipeline (10-metric framework)
• Data preprocessing scripts (cropping, resizing, normalization)
• Dataset-adaptive normalization code
• Baseline DenseNet-121 training scripts
• Classification schemes: 2-class, 3-class, 5-class
• Documentation and usage examples

Citation

If you use this dataset or code, please cite: [Paper reference will be added upon acceptance]

Requirements

• Python 3.8+
• TensorFlow 2.x / PyTorch
• OpenCV, NumPy, scikit-learn
• See requirements.txt for full dependencies

Contact

For questions or issues, please open a GitHub issue or contact - kush3489t@gmail.com