Efficient LiDAR Reflectance Compression via Scanning Serialization

TL;DR

SerLiC introduces a novel serialization-based neural compression method for LiDAR reflectance data, achieving significant volume reduction and efficiency improvements suitable for real-time applications.

Contribution

The paper presents SerLiC, a new framework that transforms 3D LiDAR data into sequences for effective neural compression, outperforming existing methods in compression ratio and speed.

Findings

Over 2x reduction in reflectance data volume

Up to 22% fewer bits than state-of-the-art methods

Lightweight version achieves >10 fps with 111K parameters

Abstract

Reflectance attributes in LiDAR point clouds provide essential information for downstream tasks but remain underexplored in neural compression methods. To address this, we introduce SerLiC, a serialization-based neural compression framework to fully exploit the intrinsic characteristics of LiDAR reflectance. SerLiC first transforms 3D LiDAR point clouds into 1D sequences via scan-order serialization, offering a device-centric perspective for reflectance analysis. Each point is then tokenized into a contextual representation comprising its sensor scanning index, radial distance, and prior reflectance, for effective dependencies exploration. For efficient sequential modeling, Mamba is incorporated with a dual parallelization scheme, enabling simultaneous autoregressive dependency capture and fast processing. Extensive experiments demonstrate that SerLiC attains over 2x volume reduction against the original reflectance data, outperforming the state-of-the-art method by up to 22% reduction of compressed bits while using only 2% of its parameters. Moreover, a lightweight version of SerLiC achieves > 10 fps (frames per second) with just 111K parameters, which is attractive for real-world applications.