# VoxelKP: A Voxel-based Network Architecture for Human Keypoint Estimation in LiDAR Data

Source: [VoxelKP: A Voxel-based Network Architecture for Human Keypoint Estimation in LiDAR Data](<https://www.sota2.com/research/paper/68fa71721e209131cea514c0>)

Collection index: [llms.txt](<https://www.sota2.com/research/llms.txt>)

Data updated: 2026-01-04T04:00:10.484Z

Authors: Jian Shi, Peter Wonka

Published: 2023-12-11T23:50:14.000Z

## Abstract

We present \\textit{VoxelKP}, a novel fully sparse network architecture tailored for human keypoint estimation in LiDAR data. The key challenge is that objects are distributed sparsely in 3D space, while human keypoint detection requires detailed local information wherever humans are present. We propose four novel ideas in this paper. First, we propose sparse selective kernels to capture multi-scale context. Second, we introduce sparse box-attention to focus on learning spatial correlations between keypoints within each human instance. Third, we incorporate a spatial encoding to leverage absolute 3D coordinates when projecting 3D voxels to a 2D grid encoding a bird's eye view. Finally, we propose hybrid feature learning to combine the processing of per-voxel features with sparse convolution. We evaluate our method on the Waymo dataset and achieve an improvement of $27\\%$ on the MPJPE metric compared to the state-of-the-art, \\textit{HUM3DIL}, trained on the same data, and $12\\%$ against the state-of-the-art, \\textit{GC-KPL}, pretrained on a $25\\times$ larger dataset. To the best of our knowledge, \\textit{VoxelKP} is the first single-staged, fully sparse network that is specifically designed for addressing the challenging task of 3D keypoint estimation from LiDAR data, achieving state-of-the-art performances. Our code is available at \\url{https://github.com/shijianjian/VoxelKP}.

## Recorded benchmark results

A research result applies to its stated dataset, split, metric and evaluation configuration. Compare like-for-like settings, preserve the metric's unit and direction, and distinguish paper-reported results from independent reproduction. The best result recorded here is not a claim of exhaustive, current global coverage.

- [3D Human Keypoint Estimation on Waymo 1.4.2](<https://www.sota2.com/research/sota/3d-human-keypoint-estimation-on-waymo-1-4-2.md>): 3D Human Keypoint Estimation; Waymo 1.4.2; MPJPE: 8.87.

- [3D Human Keypoint Estimation on Waymo](<https://www.sota2.com/research/sota/3d-human-keypoint-estimation-on-waymo.md>): 3D Human Keypoint Estimation; Waymo; MPJPE: not recorded.

- [3D Human Keypoint Estimation on Waymo 1.3.2](<https://www.sota2.com/research/sota/3d-human-keypoint-estimation-on-waymo-1-3-2.md>): 3D Human Keypoint Estimation; Waymo 1.3.2; MPJPE: not recorded.

- [3D Human Keypoint Estimation on Internal dataset Waymo](<https://www.sota2.com/research/sota/3d-human-keypoint-estimation-on-internal-dataset-waymo.md>): 3D Human Keypoint Estimation; Internal dataset Waymo; MPJPE: not recorded.

## Original sources and implementation

[Original paper](<https://arxiv.org/abs/2312.08871>)

[PDF](<https://arxiv.org/pdf/2312.08871.pdf>)

[Implementation](<https://github.com/shijianjian/VoxelKP>)