Paper ThBT17.4
Gao, Yuan (Xi'an Jiaotong University), Li, Yaochen (Xi'an Jiaotong University), Zhang, Tengwen (Xi'an Jiaotong University), Qiu, Chao (Xi'an Jiaotong University), Liu, Yuehu (Institute of Artificial Intelligence and Robotics, Xi'an Jiaoton)
Unsupervised Alignment of Traffic Elements for Domain Adaptive Semantic Segmentation
Scheduled for presentation during the Poster Session "Perception - Semantic segmentation" (ThBT17), Thursday, September 26, 2024,
14:30−16:30, Foyer
2024 IEEE 27th International Conference on Intelligent Transportation Systems (ITSC), September 24- 27, 2024, Edmonton, Canada
This information is tentative and subject to change. Compiled on December 26, 2024
|
|
Keywords Traffic Theory for ITS, Network Modeling, Sensing, Vision, and Perception
Abstract
Unsupervised domain adaptation has been used to reduce the domain shift, which would improve the performance of semantic segmentation on unlabeled real-world data. However, existing methods do not cater to the specific characteristics of traffic scene elements, leading to suboptimal alignment outcomes. In this paper, we propose a novel domain adaptation method for semantic segmentation of road scenes via unsupervised alignment of traffic elements. Firstly, a self-training framework is developed, which distinguishes the differential features between dynamic and static traffic elements, providing more accurate alignment training. Then, we designed a dynamic and static traffic elements alignment module to achieve cross-domain feature matching between the source and the target domain images. The cosine similarity maximization is applied to the alignment of dynamic traffic elements, while the prototype learning is utilized for the static traffic elements. Furthermore, the element alignment loss functions for dynamic and static traffic elements are designed to optimize the alignment modules. The experimental results demonstrate that the proposed method is superior to the existing methods on GTA5-to-Cityscapes task, which is applicable to semantic segmentation of road scenes.
|
|