One BMIPL’s paper accepted to NIPS 2018

Activities September 11, 2018


One paper of Bio-Medical Image Processing Lab.(BMIPL), “Training deep learning based denoisers without ground truth data”, has been accepted to NIPS 2018 (Annual Conference on Neural Information Processing Systems), one of the top conferences in machine learning. This work was done by Shakarim Soltanayev, a second year MS student, and Prof. Se Young Chun.

Deep learning based denoisers have yielded great performance, but they do require clean images as ground truth data. In this work, Stein’s Unbiased Risk Estimator (SURE) based training approach has been developed and it has been demonstrated that excellent performance can be achieved without ground truth data if images are contaminated by Gaussian noise.