Representation Learning for Point Cloud Understanding
It addresses the challenge of limited 3D data for applications like autonomous driving and robotics, though it appears incremental by building on existing 2D-to-3D transfer learning approaches.
This dissertation tackles the problem of improving 3D point cloud understanding by integrating pre-trained 2D models into 3D network training, resulting in significant enhancements without relying on 2D data transformation.
With the rapid advancement of technology, 3D data acquisition and utilization have become increasingly prevalent across various fields, including computer vision, robotics, and geospatial analysis. 3D data, captured through methods such as 3D scanners, LiDARs, and RGB-D cameras, provides rich geometric, shape, and scale information. When combined with 2D images, 3D data offers machines a comprehensive understanding of their environment, benefiting applications like autonomous driving, robotics, remote sensing, and medical treatment. This dissertation focuses on three main areas: supervised representation learning for point cloud primitive segmentation, self-supervised learning methods, and transfer learning from 2D to 3D. Our approach, which integrates pre-trained 2D models to support 3D network training, significantly improves 3D understanding without merely transforming 2D data. Extensive experiments validate the effectiveness of our methods, showcasing their potential to advance point cloud representation learning by effectively integrating 2D knowledge.