UWF-Net

This is the official code of the paper:

Automatic Ultra-Wide-Field Fundus Image Enhancement for Improved Clinical Diagnosis

Train

Dataset preparation

1. The dataset including FDUWI-1, FDUWI-2 subset can be accessed by sending a request to [email protected] with detailed reasons for usage.
2. Organize dataset within the following manner:

   |--datadir
       |--train
           |--A (low quality image_dir)
              photo1.jpg
              photo1.jpg
              ...
           |--B ( quality image_dir)
              photo1.jpg
              photo1.jpg
              ...
   

Pre-trained model preparation

Prepare your pre-trained disease classification model and Change the model path of it in Line 200 of models.py

Train your own model

Run train.py with

cd ./UWF-Net
python train.py --dataroot "HERE/IS/YOUR/DATADIR/" \
--save_dir "MODEL/SAVEDIR/"

Test

Model preparation

FIQA

1. See here for model downloading.
2. Change the path of FIQA model in Line 22 of fiqa.py

UWFQA

1. Download UWFQA model from the release.
2. Change the path of UWFQA model in Line 34 of uwfqa.py

Get FIQA and UWFQA score

Run test.sh with

bash test.sh