DetVPCC: RoI-based Point Cloud Sequence Compression for 3D Object Detection

TL;DR

DetVPCC tackles the poor bitrate-accuracy trade-off of VPCC for 3D object detection by introducing ROI aware encoding in a point cloud sequence pipeline. It pairs a lightweight GMM-based ROI detector with macroblock level quality control in the VPCC geometry images, enabling higher fidelity in RoIs while compressing background regions more aggressively. Across nuScenes with CenterPoint and BEVFusion-Lidar back-ends, DetVPCC consistently improves detection metrics at comparable bitrates, with larger gains for smaller or more challenging objects. The work demonstrates that ROI aware encoding can substantially reduce bandwidth requirements for machine-driven 3D perception, with practical implications for edge-cloud streaming and storage of point cloud data.

Abstract

While MPEG-standardized video-based point cloud compression (VPCC) achieves high compression efficiency for human perception, it struggles with a poor trade-off between bitrate savings and detection accuracy when supporting 3D object detectors. This limitation stems from VPCC's inability to prioritize regions of different importance within point clouds. To address this issue, we propose DetVPCC, a novel method integrating region-of-interest (RoI) encoding with VPCC for efficient point cloud sequence compression while preserving the 3D object detection accuracy. Specifically, we augment VPCC to support RoI-based compression by assigning spatially non-uniform quality levels. Then, we introduce a lightweight RoI detector to identify crucial regions that potentially contain objects. Experiments on the nuScenes dataset demonstrate that our approach significantly improves the detection accuracy. The code and demo video are available in supplementary materials.

Figures (8)

• Figure 1: The first row demonstrates the accuracy improvement of 3D object detection after applying RoI encoding (objects detected are colored in green). The second row shows the accuracy-bitrate trade-off curve of DetVPCC and VPCC when supporting CenterPoint yin2021center and BEVFusion-Lidar liu2023bevfusion.
• Figure 2: Overview of DetVPCC.
• Figure 3: Illustration of RoI Encoding. The right image compares point clouds with RoI and non-RoI encoding. The red points represent the points that appear in the non-RoI encoded frame but not in the RoI encoded frame, and blue represents the contrary. Black points represent common parts.
• Figure 4: Design of the GMM-based RoI detector.
• Figure 5: Illustration of the evaluation metric.