DG-PIC: Domain Generalized Point-In-Context Learning for Point Cloud Understanding This paper introduces DG-PIC, a novel method for domain generalized point cloud understanding that handles multiple domains and tasks within a unified model. The method combines domain generalization (DG) and point-in-context (PIC) learning to improve the generalizability of point cloud models across various tasks and domains. DG-PIC employs dual-level source prototype estimation and dual-level test-time feature shifting to align test data with source domains, without requiring any model updates during testing. The method is evaluated on a new benchmark consisting of 30,954 point cloud samples from four distinct datasets, including two synthetic and two real-world datasets. Comprehensive experiments show that DG-PIC outperforms state-of-the-art techniques significantly in three different tasks: point cloud reconstruction, denoising, and registration. The method is designed to handle unseen domains and multiple tasks within a unified model, making it suitable for practical multi-domain and multi-task settings. The paper also introduces a new benchmark for evaluating performance in this setting. The results demonstrate that DG-PIC achieves state-of-the-art performance on three different tasks.DG-PIC: Domain Generalized Point-In-Context Learning for Point Cloud Understanding This paper introduces DG-PIC, a novel method for domain generalized point cloud understanding that handles multiple domains and tasks within a unified model. The method combines domain generalization (DG) and point-in-context (PIC) learning to improve the generalizability of point cloud models across various tasks and domains. DG-PIC employs dual-level source prototype estimation and dual-level test-time feature shifting to align test data with source domains, without requiring any model updates during testing. The method is evaluated on a new benchmark consisting of 30,954 point cloud samples from four distinct datasets, including two synthetic and two real-world datasets. Comprehensive experiments show that DG-PIC outperforms state-of-the-art techniques significantly in three different tasks: point cloud reconstruction, denoising, and registration. The method is designed to handle unseen domains and multiple tasks within a unified model, making it suitable for practical multi-domain and multi-task settings. The paper also introduces a new benchmark for evaluating performance in this setting. The results demonstrate that DG-PIC achieves state-of-the-art performance on three different tasks.