[MICCAI' 22] Intervention & Interaction Federated Abnormality Detection with Noisy Clients
By Xinyu Liu
Check FCOS for installation instructions.
Step 1: Download the GLRC dataset as well as the box annotations from this URL, split the GLRC subset and convert to VOC format. Use the provided client split to replace ImageSets dir.
[DATASET_PATH]
└─ GLRC
└─ Annotations
└─ ImageSets
└─ JPEGImages
Step 2: Generating class-conditioned noisy annotations with this script or instance-dependant noise with this script
Step 3: change the data root for your dataset at paths_catalog.py.
Train with FedInI: (Our code currently only supports single-GPU training.)
python tools/train_net.py --config ./configs/federated/glrc.yaml SOLVER.ANNOTATIONS 0.3 OUTPUT_DIR output_fedini SOLVER.METHOD att
As a comparison, train with FedAvg:
python tools/train_net.py --config ./configs/federated/glrc.yaml SOLVER.ANNOTATIONS 0.3 OUTPUT_DIR output_fedavg SOLVER.METHOD ori
If you think this work is helpful for your project, kindly give it a star and citation:
@inproceedings{liu2022intervention,
title={Intervention \& Interaction Federated Abnormality Detection with Noisy Clients},
author={Liu, Xinyu and Li, Wuyang and Yuan, Yixuan},
booktitle={International Conference on Medical Image Computing and Computer-Assisted Intervention},
pages={309--319},
year={2022},
organization={Springer}
}
The work is based on FCOS.
If you have any problems, please feel free to contact me at xliu423-c@my.cityu.edu.hk. Thanks.