PoseBench3D
收藏资源简介:
PoseBench3D是一个为3D人体姿态估计而设计的标准化测试环境,旨在评估模型在多个数据集上的性能。该框架整合了四个常用的人体姿态估计数据集:H36M、GPA、3DPW和SURREAL。通过统一的接口,PoseBench3D提供了预先配置且易于修改的格式,确保与各种模型架构的兼容性。框架对18种方法进行了重新评估,并使用MPJPE和PA-MPJPE指标报告了超过100个新的跨数据集评估结果。此外,PoseBench3D还分析了各种预处理技术和数据集准备参数对模型泛化能力的影响。
PoseBench3D is a standardized test environment designed for 3D human pose estimation, aiming to evaluate model performance across multiple datasets. This framework integrates four widely-used human pose estimation datasets: H36M, GPA, 3DPW and SURREAL. Through a unified interface, PoseBench3D provides pre-configured and easily modifiable formats to ensure compatibility with various model architectures. The framework re-evaluates 18 existing methods and reports over 100 new cross-dataset evaluation results using the MPJPE and PA-MPJPE metrics. Additionally, PoseBench3D also analyzes the impact of diverse preprocessing techniques and dataset preparation parameters on model generalization ability.

- 1PoseBench3D: A Cross-Dataset Analysis Framework for 3D Human Pose Estimation加利福尼亚大学欧文分校信息与计算机科学学院 · 2025年



