PCCN:POINT Cloud Colorization Network

Abstract

This paper investigates into the colorization problem which generates color value to point clouds. It is difficult to apply image colorization methods to three dimension directly. Currently, Generative Adversarial Network (GAN) is the most reliable solution to the colorization problem. However, it cannot directly process point cloud data with geometric data structures. In this paper, we propose a scheme for point cloud colorization which can effectively solve the permutation invariance of points as well as produce realistic color effect. In order to achieve a better result, a well-designed network that aggregates the point features as global feature is built to guide color generation. After computing the global point cloud feature vector, we integrates both the local and global information by concatenating the global feature with each of the point features. Extensive experimental results demonstrate the effectiveness of the proposed algorithm.