Domain Adaptive Semantic Segmentation by Optimal Transport

Scene segmentation is widely used in autonomous driving for environmental perception. Semantic scene (3S) has gained considerable attention owing to its rich semantic information. It assigns labels the pixels an image, thereby enabling automatic image labeling. Current approaches are based mainly on convolutional neural networks (CNN), however, they rely numerous labels. Therefore, use of a small amount labeled data achieve become increasingly important. In this study, we developed domain adaptation (DA) framework optimal transport (OT) and mechanism address issue. Specifically, first generated output space via CNN superior feature representation. Second, utilized OT more robust alignment source target domains space, where plan defined well improve model. particular, reduced number network parameters made interpretable. Third, better describe multiscale properties features, constructed perform adaptation. Finally, verify performance proposed method, conducted experiment compare method with three benchmark four SOTA methods using datasets. The mean intersection-over-union (mIOU) was significantly improved, visualization results under multiple scenarios also show that performed than methods.