The project web for "Evaluation of Retinal Image Quality Assessment Networks in Different Color-spaces" in MICCAI 2019.

Fundus Enhancement: We also have a related work for "Modeling and Enhancing Low-quality Retinal Fundus Images" in IEEE TMI, 2021. The code is released in Github: Cofe-Net

Eye-Quality (EyeQ) Assessment Dataset is a re-annotatation subset from the EyePACS dataset for fundus image quality assessment.

EyeQ dataset has 28,792 retinal images with a three-level quality grading (i.e., 'Good', 'Usable' and 'Reject').

1. The original fundus images could be downloaded from EyePACS dataset.
2. All the original fundus images should be pre-porcessed by 'EyeQ_process_main.py' in folder './EyeQ_preprocess'.
3. The quality label is in './data' folder, where the 'Label_EyeQ_train.csv' and 'Label_EyeQ_test.csv' are divided by EyePACS, and the 'DR_grade' label is also from EyePACS.
4. We also release our Multiple Color-space Fusion Network (MCF-Net) based on ResNet121 in './MCFNet' folder.

Note: The trained model of MCF-Net 'DenseNet121_v3_v1.tar' (~112MB) could be download from OneDrive.

If you use this dataset and code, please cite the following papers:

[1] Huazhu Fu, Boyang Wang, Jianbing Shen, Shanshan Cui, Yanwu Xu, Jiang Liu, Ling Shao, "Evaluation of Retinal Image Quality Assessment Networks in Different Color-spaces", in MICCAI, 2019. [PDF] Note: The corrected accuracy score of MCF-Net is 0.8800.

The code is released under the Creative Commons Attribution-NonCommercial-ShareAlike 4.0 International Public License for NonCommercial use only. Any commercial use should get formal permission first.

Update log:

• 21.12.29: Added link for fundus enhancement project.
• 20.06.18: Corrected the accuracy score.
• 19.11.15: Released the code of MCF-Net.
• 19.07.10: Released the dataset.