Code for the paper nnMobileNet: Rethinking CNN for Retinopathy Research
We will appreciate any suggestions and comments. If you find this code helpful, please cite our papers. Thanks!
@misc{zhu2024nnmobilene,
title={nnMobileNe: Rethinking CNN for Retinopathy Research},
author={Wenhui Zhu and Peijie Qiu and Xiwen Chen and Xin Li and Natasha Lepore and Oana M. Dumitrascu and Yalin Wang},
year={2024},
eprint={2306.01289},
archivePrefix={arXiv},
primaryClass={eess.IV}
}
Abstract Over the past few decades, convolutional neural networks (CNNs) have been at the forefront of the detection and tracking of various retinal diseases (RD). Despite their success, the emergence of vision transformers (ViT) in the 2020s has shifted the trajectory of RD model development. The leading-edge performance of ViT-based models in RD can be largely credited to their scalability—their ability to improve as more parameters are added. As a result, ViT-based models tend to outshine traditional CNNs in RD applications, albeit at the cost of increased data and computational demands. ViTs also differ from CNNs in their approach to processing images, working with patches rather than local regions, which can complicate the precise identification of small, variably presented lesions in RD. In our study, we revisited and updated the architecture of a CNN model, specifically MobileNet, to enhance its utility in RD diagnostics. We found that an optimized MobileNet, through selective modifications, can surpass ViT-based models in various RD benchmarks, including diabetic retinopathy grading, detection of multiple fundus diseases, and classification of diabetic macular edema.
We also brought this model to MICCAI MMAC 2023, and won the 3rd.
---------------------------------------------------------------------------------------------
In main.py -> ImageNet pretrained weight
model = ReXNetV1(width_mult=3.0,classes=args.nb_classes,dropout_path=args.drop_path)
model.load_state_dict(torch.load('rexnet_3.0.pth'),strict=False)
---------------------------------------------------------------------------------------------
rexnet_3.0.pth download from Google Drive Or Rexnet Official repo
Single-GPU
python main.py --input_size 224 --drop_path 0.2
Multi-GPU
python -m torch.distributed.launch --nproc_per_node=4 main.py --batch_size 24 --lr 1e-3 --update_freq 1 --weight_decay 0.005 --min_lr 1e-6 --input_size 512 --drop_path 0.21 --data_set MICCAI --nb_classes 5 --warmup_epochs 20 --epochs 1000 --main_eval avarage --cutmix 1.0 --opt adamp --mixup 0.8 --log_dir Experiment/size_512_dp_0.21_DataArg_True/fold_5/log --output_dir Experiment/size_512_dp_0.21_DataArg_True/fold_5 --fold_train kfold_csv/train_fold_5.csv --fold_test kfold_csv/test_fold_5.csv
Released under the ASU GitHub Project License.
