Att2CPC: Attention-Guided Lossy Attribute Compression of Point Clouds

TL;DR

This article proposes an efficient attention-based method for lossy compression of point cloud attributes leveraging on an autoencoder architecture and is the first approach to introduce attention mechanism to point-based lossy PCAC task.

Abstract

With the great progress of 3D sensing and acquisition technology, the volume of point cloud data has grown dramatically, which urges the development of efficient point cloud compression methods. In this paper, we focus on the task of learned lossy point cloud attribute compression (PCAC). We propose an efficient attention-based method for lossy compression of point cloud attributes leveraging on an autoencoder architecture. Specifically, at the encoding side, we conduct multiple downsampling to best exploit the local attribute patterns, in which effective External Cross Attention (ECA) is devised to hierarchically aggregate features by intergrating attributes and geometry contexts. At the decoding side, the attributes of the point cloud are progressively reconstructed based on the multi-scale representation and the zero-padding upsampling tactic. To the best of our knowledge, this is the first approach to introduce attention mechanism to point-based lossy PCAC task. We verify the compression efficiency of our model on various sequences, including human body frames, sparse objects, and large-scale point cloud scenes. Experiments show that our method achieves an average improvement of 1.15 dB and 2.13 dB in BD-PSNR of Y channel and YUV channel, respectively, when comparing with the state-of-the-art point-based method Deep-PCAC. Codes of this paper are available at https://github.com/I2-Multimedia-Lab/Att2CPC.

Figures (10)

• Figure 1: The overall framework of our approach. ECA represents the External Cross Attention Module that abstracts color features based on geometry context. FPS denotes the Farthest Point Sampling that extracts uniform sampling points from the original point cloud. AE and AD refers to arithmetic encoding and arithmetic decoding, respectively.
• Figure 2: Structure of External Cross Attention module (ECA).
• Figure 3: Structure of Internal Self-Attention module (ISA).
• Figure 4: Upsampling Block. The black points indicates that the attribute value of the points is to be recontrusted, and the white points indicates that the attribute value of the points is zero. $\mathcal{P}_s$ and $\mathcal{F}_s$ denotes the input point geometry coordinates and input features, respectively. $\mathcal{P}_{s+1}$ and $\mathcal{F}_{s+1}$ denotes the upsampled coordinates and features, respectively.
• Figure 5: Examples of our training set. Point clouds are synthesized from ShapeNet shapenet and PCCD pccd.