LiDAR-PTQ: Post-Training Quantization for Point Cloud 3D Object Detection

TL;DR

LiDAR-PTQ addresses the challenge of deploying LiDAR-based 3D object detectors on edge devices by tailoring post-training quantization to sparse point clouds. It introduces a three-component framework: sparsity-based calibration to initialize quantization parameters, a Task-guided Global Positive Loss (TGPL) to align quantized outputs with the FP model in a task-aware way, and adaptive rounding-to-nearest to minimize layer-wise reconstruction error. Across CenterPoint Pillar and Voxel variants, LiDAR-PTQ achieves INT8 accuracy that closely matches FP32 while delivering about 3× inference speedups on edge hardware and approximately 30× faster quantization than QAT, with strong results on both sparse and dense detectors. These findings suggest LiDAR-PTQ as a practical, generalizable tool for efficient, deployable 3D detectors on edge platforms, enabling robust performance with reduced data and compute requirements.

Abstract

Due to highly constrained computing power and memory, deploying 3D lidar-based detectors on edge devices equipped in autonomous vehicles and robots poses a crucial challenge. Being a convenient and straightforward model compression approach, Post-Training Quantization (PTQ) has been widely adopted in 2D vision tasks. However, applying it directly to 3D lidar-based tasks inevitably leads to performance degradation. As a remedy, we propose an effective PTQ method called LiDAR-PTQ, which is particularly curated for 3D lidar detection (both SPConv-based and SPConv-free). Our LiDAR-PTQ features three main components, \textbf{(1)} a sparsity-based calibration method to determine the initialization of quantization parameters, \textbf{(2)} a Task-guided Global Positive Loss (TGPL) to reduce the disparity between the final predictions before and after quantization, \textbf{(3)} an adaptive rounding-to-nearest operation to minimize the layerwise reconstruction error. Extensive experiments demonstrate that our LiDAR-PTQ can achieve state-of-the-art quantization performance when applied to CenterPoint (both Pillar-based and Voxel-based). To our knowledge, for the very first time in lidar-based 3D detection tasks, the PTQ INT8 model's accuracy is almost the same as the FP32 model while enjoying $3\times$ inference speedup. Moreover, our LiDAR-PTQ is cost-effective being $30\times$ faster than the quantization-aware training method. Code will be released at \url{https://github.com/StiphyJay/LiDAR-PTQ}.

Figures (3)

• Figure 1: The sparsity of point cloud on 3D LiDAR-based object detection. Orange area means empty area, blue point means the point cloud (non-empty area) in a scenario, green box means the 3D Bboxes, and red point means foreground points.
• Figure 2: Performance comparison
• Figure 3: The diagram of data distribution for RGB-based and LiDAR-based object detection. Orange and green denote the data distribution of the entire feature map and foreground feature.