We propose a method for generating high-quality pseudo-masks using CAM obtained through a network with large kernels. Our network achieved a 23.09% mIoU improvement over ResNet-18 and 7.55% over VGG-16. Moreover, through qualitative assessments, our network demonstrated high-quality pseudo-masks.

[speaker]
ARo Yun is currently a master's course student in the Department of Imaging Engineering in the Graduate School of Advanced Imaging Science, Multimedia & Film at Chung-Ang University. Her research interests include deep learning and computer vision.