EfficientQ: An efficient and accurate post-training neural network quantization method for medical image segmentation

Rongzhao Zhang, Albert C. S. Chung

Abstract: Highlights•An efficient quantization method is designed for medical image segmentation models.•The problem is solved using ADMM, with a prior-aware and self-adaptive objective.•A UNet can be quantized to 2-bit with minimal performance degradation in 5 minutes.•Performance is verified on the public LiTS and BraTS datasets.