目前有两篇学术中共有两篇论文以不同的形式提出了boundary loss,分别是论文1: Boundary Loss for Remote Sensing Imagery Semantic Segmentation 与论文2:Boundary loss for highly unbalanced segmentation 。
论文1所提出的boundary loss即最小化label边缘与pred边缘的f-score(也就是dice loss),其项目地址如下所示。代码为pytorch版本,稍后博主会分享改写的paddle版本代码。GitHub - yiskw713/boundary_loss_for_remote_sensing: Pytorch re-implementation of boundary loss, proposed in "Boundary Loss for Remote Sensing Imager