QFedPolyp: Federated Learning for Polyp Segmentation

QFedPolyp is an efficient framework for federated learning aimed at collaborative polyp segmentation, focusing on effective communication and inference. It integrates quantization-aware training with low-precision model exchanges to minimize communication expenses while ensuring privacy. Hospitals conduct local training of a lightweight U-Net using their private datasets, simulating quantization throughout the training process. The quantized model parameters are then sent to a central server for aggregation through Federated Averaging. The evaluation utilizes datasets including Kvasir-SEG, CVC-ClinicVideoDB, PolypGen, and BKAI-IGH NeoPolyp.