Efficient and Generic Point Model for Lossless Point Cloud Attribute Compression

TL;DR

The paper tackles lossless point cloud attribute compression (PCAC) for diverse point clouds, addressing high computational cost and poor generalization in prior learning-based methods. It introduces PoLoPCAC, a point-model framework that operates directly on raw PCG anchors and models lossless PCAC as inferring explicit distributions from inter-group autoregressive priors $P_{ heta}(Y)$, with a group-wise factorization across progressively grouped blocks. A Progressive Random Grouping (PRG) strategy partitions points into groups of increasing size, enabling inter-group autoregressive coding and intra-group parallel coding, while a locality-aware attention network estimates per-point distribution parameters from adaptive $K$NN context windows. The approach achieves competitive bitrate performance with a small model size (~2.6MB) and faster coding times than G-PCCv23 on many sequences, and generalizes well across ShapeNet, ScanNet, MVUB, and 8iVFB after training on a synthetic dataset (Synthetic 2k-ShapeNet). The authors provide code, dataset, and models, highlighting practical applicability for scalable, distortion-free lossless PCAC.

Abstract

The past several years have witnessed the emergence of learned point cloud compression (PCC) techniques. However, current learning-based lossless point cloud attribute compression (PCAC) methods either suffer from high computational complexity or deteriorated compression performance. Moreover, the significant variations in point cloud scale and sparsity encountered in real-world applications make developing an all-in-one neural model a challenging task. In this paper, we propose PoLoPCAC, an efficient and generic lossless PCAC method that achieves high compression efficiency and strong generalizability simultaneously. We formulate lossless PCAC as the task of inferring explicit distributions of attributes from group-wise autoregressive priors. A progressive random grouping strategy is first devised to efficiently resolve the point cloud into groups, and then the attributes of each group are modeled sequentially from accumulated antecedents. A locality-aware attention mechanism is utilized to exploit prior knowledge from context windows in parallel. Since our method directly operates on points, it can naturally avoids distortion caused by voxelization, and can be executed on point clouds with arbitrary scale and density. Experiments show that our method can be instantly deployed once trained on a Synthetic 2k-ShapeNet dataset while enjoying continuous bit-rate reduction over the latest G-PCCv23 on various datasets (ShapeNet, ScanNet, MVUB, 8iVFB). Meanwhile, our method reports shorter coding time than G-PCCv23 on the majority of sequences with a lightweight model size (2.6MB), which is highly attractive for practical applications. Dataset, code and trained model are available at https://github.com/I2-Multimedia-Lab/PoLoPCAC.

Figures (11)

• Figure 1: PoLoPCAC. Input point cloud is first resolved into several groups (left); Then, attributes of each group are predicted sequentially using accumulated antecedents as context (middle); A locality-aware attention mechanism is utilized to infer the probability distribution of attributes from context windows in parallel (right).
• Figure 2: An illustration of fast grouping strategy for practical implementation.
• Figure 3: Operational diagram for our group-wize autoregressive modeling. Each group is modeled serially by using previous groups as context.
• Figure 4: Attention mechanism used in this work. query of the target point is masked to zero thus key is directly utilized to fuse positional encodings. pem refers to the position encoding multiplier. Superscript is omitted for a concise explanation.
• Figure 5: Datasets used in this paper. We limit the training on the Synthetic 2k-ShapeNet training set and test on a variety of datasets.