Geometric Representation Learning for 3D Understanding: From Point Clouds to Gaussian Splatting
About this Topic:
Point cloud classification is a fundamental yet persistently challenging task in three-dimensional data analysis. This presentation introduces a network designed to capture the geometric structure of point clouds for richer and more discriminative representations. The approach enriches low-level geometric information among points explicitly, while convolution-based structures in higher-level feature spaces learn local geometric context implicitly. Central to the method is an error-correcting feedback mechanism that captures local features comprehensively, complemented by a channel-affinity attention module that reduces redundancy by emphasizing the most distinctive feature channels. Experiments on both synthetic and real-world datasets demonstrate that the method achieves a strong balance between accuracy and efficiency compared with leading alternatives. Building on this foundation, the talk extends the discussion toward more recent advances in geometric representation learning for three-dimensional recognition and understanding, including emerging Gaussian representations and applications to specialized domains such as medical data analysis.
About the Presenter:
Shi Qiu (M’24) received the Ph.D. degree in engineering and computer science from The Australian National University in 2023.
He is currently a Research Assistant Professor with the Department of Computer Science and Engineering, The Chinese University of Hong Kong, where he is also affiliated with the Institute of Medical Intelligence and XR. His research interests include 3D computer vision, graphics, visualization, and extended reality, with a particular focus on medical AIXR applications.
Dr. Qiu’s work has been published in leading venues such as IEEE T-PAMI, T-MM, CVPR, ICCV, IEEE VR, and IEEE VIS.
Want to learn more about upcoming events & webinars?
Visit the events section of the Signal Processing website to see all upcoming lectures, workshops, webinars, and more.