Code for deep learning on color fundus photographs for early recognition and differential diagnosis of retinal vascular occlusion.
Notification
- Here we provide the interface of the DeepDrRVO framework and its submodules (DeepDrVAN, DeepDrVBC, DeepDrABC) by pytorch for early recognition and differential diagnosis of retinal vascular occlusion.
- Code for the training of the individual modules and the Few-Sample Generator (FSG) is available from the corresponding author upon reasonable request.
pip install -r requirements.txt
WMUEH is available from the corresponding author upon reasonable request.
RFMiD Retinal fundus multi-disease image dataset.
ODIR Ocular Disease Intelligent Recognition.
JSIEC.
If you want to get the synthetic CFPs of BRAO or CRAO by Few-Sample Generator please run ./tools/generation.py
after setting the args
.
- Setting number of generated synthetic CFPs at
--num
. - Setting generate CPF type at
--RaO
. - Setting synthetic CFPs storage directory at
--save_path
. - Run
./tools/generation.py
.
We provide the .pth
file of the optimal model for each module at Baidu pan.
Through the main.py
, you can start using the modules of DeepDrRVO to realize early recognition and differential diagnosis of retinal vascular occlusion.
Please follow these steps before using in your own dataset. Or use the sample we provided for demonstration.
- Setting work mode at
--module
. - Setting CFPs path at
--image_root_dir
. - Setting batch size at
--batch_size
. - Run
main.py
.
When the inference is completed, the results will be stored in ./results.csv
.