Deep point cloud compression coding method based on full self-attention network

Abstract

The invention discloses a depth point cloud compression coding method based on a full-self-attention network, and the method comprises the steps: constructing a point cloud full-self-attention network which comprises an encoder and a decoder; obtaining training data, and constructing a chamfering distance objective function to train the point cloud full-self-attention network; inputting point cloud data into the trained point cloud full-self-attention network, performing feature sampling processing on the point cloud data by using an encoder to obtain a point cloud code, and completing point cloud compression; and reconstructing point cloud data by using a decoder according to the point cloud code to complete point cloud decompression. According to the method, learning of local and global correlation among points of the point cloud is enhanced based on network training of a chamfering distance objective function, point cloud codes capable of accurately representing semantic information of the point cloud are obtained through feature sampling of the point cloud by a coding machine, safety and stability of storage and transportation of the point cloud information are guaranteed, and the method is suitable for large-scale popularization and application. The method can be widely applied to the technical field of point cloud compression coding.

Description

technical field [0001] The invention relates to the technical field of point cloud compression encoding, in particular to a deep point cloud compression encoding method based on a full self-attention network. Background technique [0002] Learning-based point cloud compression is a specific implementation method of point cloud compression. A point cloud is a collection of discrete points with 3D geometric locations and other attributes (e.g., color, opacity, etc.), which can be used to efficiently represent stereoscopic data such as 3D scenes and objects. Realizing efficient point cloud compression methods can promote the progress of various point cloud applications, such as augmented reality, autonomous driving, 3D free-viewpoint video, holographic transmission, etc. Recently, learning-based methods for point cloud compression have emerged, using powerful deep learning networks to learn the parameters of point cloud compression encoders. There are two main problems in exi...

AI Technical Summary

Problems solved by technology

One is that the method based on voxel input does not fully consider the sparsity of the point cloud, and the complexity is too high to be practically applied; the other is that the method based on point input does not make full use of the correlation between points, and the coding efficiency is low

Images