ELiC: Efficient LiDAR Geometry Compression via Cross-Bit-depth Feature Propagation and Bag-of-Encoders

TL;DR

ELiC addresses real-time LiDAR geometry compression by integrating cross-bit-depth feature propagation, a Bag-of-Encoders pool for depth-adaptive modeling, and a Morton-order-preserving hierarchical traversal. This combination enhances entropy modeling and reduces latency, delivering state-of-the-art or competitive rate–distortion performance on Ford and SemanticKITTI while maintaining real-time throughput. The approach achieves this with a compact model family and efficient encoding/decoding pipelines, offering practical deployment potential and clear avenues for future optimization, including quantization-aware training and level-dependent factorization.

Abstract

Hierarchical LiDAR geometry compression encodes voxel occupancies from low to high bit-depths, yet prior methods treat each depth independently and re-estimate local context from coordinates at every level, limiting compression efficiency. We present ELiC, a real-time framework that combines cross-bit-depth feature propagation, a Bag-of-Encoders (BoE) selection scheme, and a Morton-order-preserving hierarchy. Cross-bit-depth propagation reuses features extracted at denser, lower depths to support prediction at sparser, higher depths. BoE selects, per depth, the most suitable coding network from a small pool, adapting capacity to observed occupancy statistics without training a separate model for each level. The Morton hierarchy maintains global Z-order across depth transitions, eliminating per-level sorting and reducing latency. Together these components improve entropy modeling and computation efficiency, yielding state-of-the-art compression at real-time throughput on Ford and SemanticKITTI. Code and models will be released upon publication.

Figures (11)

• Figure 1: Average encoding and decoding FPS for 12-bit LiDAR geometry compression, and BD-Rate savings relative to G-PCC v30 across 16-, 15-, 14-, 13-, and 12-bit LiDAR geometries. Our models (ELiC and ELiC-Large) achieve real-time compression at 12-bit input (LiDAR capture speed ${\ge}10$ FPS) while maintaining compression efficiency comparable to state-of-the-art methods. Results are reported on Ford and SemanticKITTI datasets.
• Figure 2: Average number of neighboring points by coordinate resolution (bit-depth) in the Ford-01 sequence (1,500 frames).
• Figure 3: Architecture of the coding network in ELiC. The diagram shows encoding and decoding pipelines together. For the separated pipelines, see Fig. \ref{['fig:pipeline']} in the supplementary material.
• Figure 4: Concept diagram of the Bag-of-Encoders (BoE) coding strategy in ELiC. At each bit-depth level, ELiC selects a coding network from a BoE pool based on the occupancy distribution and adaptively assembles the model for LiDAR geometry compression.
• Figure 5: Per-point bit allocation on SemanticKITTI frame at the 15 and 12 bit-depth levels for RENO, ELiC w/o BoE, and ELiC ($K{=}5$). The colormap encodes lower (blue) to higher (dark orange) predicted bits per point.