/LANet

Diabetic Retinopathy Grading with Weakly-Supervised Lesion Priors

Primary LanguagePython

LANet

This repo covers the official implementation of paper DIABETIC RETINOPATHY GRADING WITH WEAKLY-SUPERVISED LESION PRIORS

Dataset

Download the DDR Dataset from DDR (github)

Train

We provide the implementation of LANet with different backbones, including resnet, densenet, vgg, mobilenet, efficientnet, inceptionv3. Take resnet as an example:

baseline model

DR recognition (binary classification)

python main_base.py --n_classes 2 --model res50 --visname ddr_res50_base2

DR grading (5-grade classification)

python main_base.py --n_classes 5 --model res50 --visname ddr_res50_base5

lanet model

LANet (w/o adaptive loss)

python main_lanet.py --model res50 --visname ddr_res50_lanet

LANet (w/ adaptive loss)

python main_lanet.py --model res50 --adaloss True --visname ddr_res50_lanet_adl

Test

baseline model

python main_base.py --dataset ddr --model res50 --visname tests --n_classes 5 --test True

LANet (w/ adaptive loss)

python main_lanet.py --dataset ddr --model res50 --visname tests --adaloss True --test True

Reference

@inproceedings{hou2023diabetic,
  title={Diabetic Retinopathy Grading with Weakly-Supervised Lesion Priors},
  author={Hou, Junlin and Xiao, Fan and Xu, Jilan and Feng, Rui and Zhang, Yuejie and Zou, Haidong and Lu, Lina and Xue, Wenwen},
  booktitle={ICASSP 2023-2023 IEEE International Conference on Acoustics, Speech and Signal Processing (ICASSP)},
  pages={1--5},
  year={2023},
  organization={IEEE}
}