DiffuSaL: Diffusion-Based Scene and Label Space Extension for Domain Generalization in Driving-Scene Semantic Segmentation
TL;DR
DiffuSaL uses diffusion models to extend scene and label spaces, addressing domain shift in driving-scene semantic segmentation for better generalization to unseen scenarios.
DiffuSaL: Diffusion-Based Scene and Label Space Extension for Domain Generalization in Driving-Scene Semantic Segmentation
Noha Nekamiche; Ibrahim Kajo; Mohamed Kas; Yassine Ruichek
https://doi.org/10.1109/ACCESS.2025.3646942
Volume 14
Generalizing to unseen scenes presents a significant challenge to semantic segmentation models, as it aims to replicate the human adaptability. Existing deep learning models still suffer from performance degradation on unseen scenarios due to the domain shift problem, which arises from differences in data statistical distribution between the training and unseen testing domains. In this work, we in...