Problem
Research questions and friction points this paper is trying to address.
Human LiDAR
Point Cloud Generation
Data Scarcity
Human Pose Estimation
HuLiGen: Human LiDAR Generation from Parametric Body Models
🤖 AI Summary
This study addresses the scarcity of real-world human LiDAR data and the domain gap inherent in simulated alternatives by proposing a pre-training scheme based exclusively on synthetic data. Methodologically, the work constructs a high-fidelity human LiDAR point cloud generation framework grounded in parametric body models, integrating a Point Transformer architecture with Flow Matching generative techniques. This approach enables high-quality pre-training without requiring real scanned data, thereby significantly enhancing downstream 3D human pose estimation performance. Experimental results demonstrate that the proposed method achieves state-of-the-art performance, reducing the Mean Per Joint Position Error (MPJPE) by up to 50%. These advantages are particularly pronounced in low-annotation regimes. The source code has been made publicly available.
📝 Abstract
LiDAR point clouds of humans are extremely expensive to collect and annotate, thus represent a scarce resource that hinders the development of human analysis using this modality. To alleviate this scarcity, prior work relies on simulated human LiDAR, but such samples do not fully reflect the geometry and sensing characteristics of real observations. In contrast, we introduce HuLiGen, a generative model that generates human LiDAR point clouds from a parametric body model, using a point transformer trained with a flow-matching objective. We show that our generated point clouds are closer to the real capture distribution. Using HuLiGen to generate synthetic data, we propose a synthetic-only pretraining scheme for LiDAR-based HPE that achieves state-of-the-art performance, with even larger gains in low-annotation and low-data regimes, where MPJPE is reduced by up to 50%. Code, models and generated samples are available at https://github.com/valeoai/HuLiGen.
Innovation
Methods, ideas, or system contributions that make the work stand out.
Human LiDAR Generation
Flow Matching
Point Transformer
Parametric Body Models
Human Pose Estimation
🔎 Similar Papers
No similar papers found.
