Due to the random growth mechanism of the seed region in the weakly supervised semantic segmentation task, the weakly supervised semantic segmentation network often suffers from wrong segmentation and missed segmentation problems. To address the above problems, this paper proposed a boundary-assisted weakly supervised semantic segmentation network. The network provided a reference for the growth of seed regions by utilizing boundary information and semantic information, so that the seed regions could naturally grow to the target boundary and correctly differentiated the target categories when the non-target object blocked or overlapped the target, and generated pseudo-pixel masks that could cover a more complete target. This paper used the pseudo-pixel mask as the supervisory information to train the segmentation network, which could improve the problem of missegmentation and omission by the pseudo-pixel mask that couldn't cover the target region accurately, and improved the accuracy of the weakly supervised semantic segmentation network. It evaluated the network on the generalized dataset PASCAL VOC 2012 validation and test sets, and the mIoU reaches 71.7% and 73.2%, respectively. The experimental results show that the performance of the proposed network outperforms most of the current weakly-supervised semantic segmentation methods at the image level.

[2023-08-22] Accepted Paper /doc/2023.06.0265/v1

[2024-02-05] Printed Article /doc/2023.06.0265/v2