The proposed Inter and Intra-Domain Mixing (IIDM) framework effectively leverages both inter-domain and intra-domain information to mitigate the domain shift and enhance the performance of semi-supervised domain adaptation in semantic segmentation.
The paper introduces a novel Convolution-based Probability Gradient (CPG) loss function that enhances the performance of semantic segmentation networks by maximizing the similarity between the predicted and ground-truth probability gradients, particularly at object boundaries.
A novel training framework, Decouple Defogging and Semantic learning (D2SL), that learns better semantics for segmentation while maintaining defogging ability by decoupling defogging and semantic learning tasks.
The core message of this article is that the established practice of hierarchical semantic segmentation may be limited to in-domain settings, whereas flat classifiers generalize substantially better, especially if they are modeled in the hyperbolic space.
ECAP proposes a novel data augmentation strategy to reduce the impact of erroneous pseudo-labels in unsupervised domain adaptive semantic segmentation.
提案されたECAPは、信頼性の高い疑似ラベルを活用することで、誤った疑似ラベルの影響を軽減し、セマンティックセグメンテーションの未監督ドメイン適応におけるパフォーマンスを向上させます。