Erroneous pixel prediction for semantic image segmentation
Erroneous pixel prediction for semantic image segmentation作者机构:State Key Lab of CAD&CGZhejiang UniversityHangzhou 310058China School of Computing Clemson UniversitySouth Carolina29634USA
出 版 物:《Computational Visual Media》 (计算可视媒体(英文版))
年 卷 期:2022年第8卷第1期
页 面:165-175页
核心收录:
学科分类:08[工学] 080203[工学-机械设计及理论] 0835[工学-软件工程] 0802[工学-机械工程]
基 金:supported by the National Natural Science Foundation of China(No.61732016)
主 题:erroneous pixel prediction image segmentation deep learning
摘 要:We consider semantic image *** method is inspired by Bayesian deep learning which improves image segmentation accuracy by modeling the uncertainty of the network *** contrast to uncertainty,our method directly learns to predict the erroneous pixels of a segmentation network,which is modeled as a binary classification *** can speed up training comparing to the Monte Carlo integration often used in Bayesian deep *** also allows us to train a branch to correct the labels of erroneous *** method consists of three stages:(i)predict pixel-wise error probability of the initial result,(ii)redetermine new labels for pixels with high error probability,and(iii)fuse the initial result and the redetermined result with respect to the error *** formulate the error-pixel prediction problem as a classification task and employ an error-prediction branch in the network to predict pixel-wise error *** also introduce a detail branch to focus the training process on the erroneous *** have experimentally validated our method on the Cityscapes and ADE20K *** model can be easily added to various advanced segmentation networks to improve their *** DeepLabv3+as an example,our network can achieve 82.88%of mloU on Cityscapes testing dataset and 45.73%on ADE20K validation dataset,improving corresponding DeepLabv3+results by 0.74%and 0.13%respectively.