The paper investigates domain adaptive semantic segmentation in foggy weather scenarios, which aims to enhance the utilization of unlabeled foggy data and improve the model's adaptability to foggy conditions. Current methods rely on clear images as references and jointly learn defogging and segmentation for foggy images, which has two main drawbacks: (1) the coupling of segmentation and defogging feature representations, resulting in a decrease in semantic representation capability, and (2) the failure to leverage real fog priors in unlabeled foggy data, leading to insufficient model generalization ability.
To address these issues, the authors propose the D2SL framework, which decouples defogging and semantic learning tasks. D2SL introduces a Domain-Consistent Transfer (DCT) strategy to establish a connection between defogging and segmentation tasks, and a Real Fog Transfer (RFT) strategy to improve defogging effects by fully leveraging the fog priors from real foggy images. This enhances the semantic representations required for segmentation during the defogging learning process and maximizes the representation capability of fog invariance by effectively utilizing real fog data.
Comprehensive experiments validate the effectiveness of the proposed method, demonstrating that D2SL consistently delivers robust performance across various domain-adaptive tasks in foggy conditions and outperforms contemporary methods.
Til et andet sprog
fra kildeindhold
arxiv.org
Vigtigste indsigter udtrukket fra
by Xuan Sun,Zha... kl. arxiv.org 04-09-2024
https://arxiv.org/pdf/2404.04807.pdfDybere Forespørgsler