# Unsupervised Domain Adaptation for Semantic Segmentation with Pseudo Label Self-Refinement Source: [Unsupervised Domain Adaptation for Semantic Segmentation with Pseudo Label Self-Refinement]() Collection index: [llms.txt]() Data updated: 2025-10-26T02:10:33.419Z Authors: Xingchen Zhao, Niluthpol Chowdhury Mithun, Abhinav Rajvanshi, Han-Pang Chiu, Supun Samarasekera Published: 2023-10-25T20:31:07.000Z ## Abstract Deep learning-based solutions for semantic segmentation suffer from significant performance degradation when tested on data with different characteristics than what was used during the training. Adapting the models using annotated data from the new domain is not always practical. Unsupervised Domain Adaptation (UDA) approaches are crucial in deploying these models in the actual operating conditions. Recent state-of-the-art (SOTA) UDA methods employ a teacher-student self-training approach, where a teacher model is used to generate pseudo-labels for the new data which in turn guide the training process of the student model. Though this approach has seen a lot of success, it suffers from the issue of noisy pseudo-labels being propagated in the training process. To address this issue, we propose an auxiliary pseudo-label refinement network (PRN) for online refining of the pseudo labels and also localizing the pixels whose predicted labels are likely to be noisy. Being able to improve the quality of pseudo labels and select highly reliable ones, PRN helps self-training of segmentation models to be robust against pseudo label noise propagation during different stages of adaptation. We evaluate our approach on benchmark datasets with three different domain shifts, and our approach consistently performs significantly better than the previous state-of-the-art methods. ## Recorded benchmark results A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage. - [Semantic Segmentation on EXPo Pq -\> Pv](): Semantic segmentation; EXPo Pq -\> Pv; Accuracy: 72.93. - [Semantic Segmentation on EXPo vehicle to drone 1.0 (test)](): Semantic segmentation; EXPo vehicle to drone 1.0 (test); Accuracy: 64.61. - [Semantic Segmentation on EXPo vehicle (Pv) to quadruped (Pq)](): Semantic segmentation; EXPo vehicle (Pv) to quadruped (Pq); Accuracy: 70.91. - [Semantic Segmentation on EXPo quadruped to vehicle v1.0](): Semantic segmentation; EXPo quadruped to vehicle v1.0; Accuracy: 72.93. - [Semantic Segmentation on EXPo Drone-to-Quadruped v1 (test)](): Semantic segmentation; EXPo Drone-to-Quadruped v1 (test); Accuracy: 67.83. - [Semantic Segmentation on EXPo drone to quadruped 1.0 (test)](): Semantic segmentation; EXPo drone to quadruped 1.0 (test); Accuracy (Overall): 67.83. - [Semantic Segmentation on EXPo Pv -\> Pd](): Semantic segmentation; EXPo Pv -\> Pd; Accuracy: 64.61. - [Semantic Segmentation on EXPo Pv -\> Pq](): Semantic segmentation; EXPo Pv -\> Pq; Acc: 70.91. - [Semantic Segmentation on EXPo Pd -\> Pq](): Semantic segmentation; EXPo Pd -\> Pq; Acc: 72.93. - [Semantic Segmentation on EXPo Average](): Semantic segmentation; EXPo Average; Accuracy: 68.72. - [Semantic Segmentation on EXPo Drone-to-Vehicle target v1 (test)](): Semantic segmentation; EXPo Drone-to-Vehicle target v1 (test); Acc: 72.46. - [Semantic Segmentation on EXPo Quadruped to Drone 1.0 (test)](): Semantic segmentation; EXPo Quadruped to Drone 1.0 (test); Accuracy: 63.57. - [Semantic Segmentation on EXPo Pq -\> Pd](): Semantic segmentation; EXPo Pq -\> Pd; Accuracy: 63.57. - [Semantic Segmentation on EXPo Pd -\> Pv](): Semantic segmentation; EXPo Pd -\> Pv; Accuracy: 67.83. ## Original sources and implementation [Original paper]() [PDF]()