Learned Nonlinear Predictor for Critically Sampled 3D Point Cloud Attribute Compression

Abstract

We study 3D point cloud attribute compression via a volumetric approach: assuming point cloud geometry is known at both encoder and decoder, parameters of a continuous attribute function f: R^3 R are quantized to and encoded, so that discrete samples f_(x_i) can be recovered at known 3D points x_i R^3 at the decoder. Specifically, we consider a nested sequences of function subspaces F^(p)_l_0 F^(p)_L, where F_l^(p) is a family of functions spanned by B-spline basis functions of order p, f_l^* is the projection of f on F_l^(p) represented as low-pass coefficients F_l^*, and g_l^* is the residual function in an orthogonal subspace G_l^(p) (where G_l^(p) F_l^(p) = F_l+1^(p)) represented as high-pass coefficients G_l^*. In this paper, to improve coding performance over do2023volumetric, we study predicting f_l+1^* at level l+1 given f_l^* at level l and encoding of G_l^* for the p=1 case (RAHT(1)). For the prediction, we formalize RAHT(1) linear prediction in MPEG-PCC in a theoretical framework, and propose a new nonlinear predictor using a polynomial of bilateral filter. We derive equations to efficiently compute the critically sampled high-pass coefficients G_l^* amenable to encoding. We optimize parameters in our resulting feed-forward network on a large training set of point clouds by minimizing a rate-distortion Lagrangian. Experimental results show that our improved framework outperforms the MPEG G-PCC predictor by 11\%--12\% in bit rate.

Tasks

Reproductions