arxivcs.CV2026-07-11
CVKD-UDA: Cross-View Knowledge Distillation for 3D Unsupervised Domain Adaptive Segmentation
Zhimin Yuan, Ming Cheng, Shangshu Yu, Wen Li, Dunqiang Liu, Xin Huang, et al.
3D unsupervised domain adaptive (UDA) segmentation mitigates the high cost of manual annotations of the new domain data. Self-training has emerged as the dominant approach in this area, where its success heavily depends on a well-initialized warm-up model to generate reliable pse…