点云
计算机科学
分割
生成模型
人工智能
生成语法
光学(聚焦)
对象(语法)
点(几何)
计算机视觉
过程(计算)
数学
几何学
操作系统
光学
物理
作者
Maciej Zamorski,Maciej Zięba,Jerzy Świątek
标识
DOI:10.1007/978-3-030-61401-0_28
摘要
The three-dimensional data representations have found numerous applications, most notably in SLAM and autonomous driving, where the most widely used type of data is a point cloud. In contrast to the image data, point clouds are unstructured objects, represented as sets of points in three- or six-dimensional (if the colors of surroundings are captured as well) space. Each of the point clouds can have a variable number of points, that in turn, are in \(\mathbb {R}^3\) space. All those factors dramatically increase the complexity of the data. The PointNet model offers an easy way to process point cloud data and can perform classification and segmentation tasks, based on raw point clouds. However, in literature, PointNet is usually trained on the complete data that captures the shape features from every side of the object. In real-world applications, the collected data may be moved, occluded, and noisy. In this work, we focus on training generative models on partial data in order to predict how the point clouds may have looked as complete objects. Such completed point clouds may then later be used in other downstream tasks.
科研通智能强力驱动
Strongly Powered by AbleSci AI