SMPL Mannequin Benchmark
收藏资源简介:
SMPL Mannequin Benchmark是由NAVER LABS Europe创建的一个包含24,428帧图像的数据集,用于评估在自然环境中的3D人体姿态估计。该数据集利用Mannequin Challenge数据集中的静态人体视频,通过结构从运动技术精确拟合SMPL模型。数据集涵盖多种身体姿态、外观和环境,包括室内外场景,自然遮挡和近距离拍摄。创建过程涉及人体实例的跟踪和分割,3D点云重建,以及SMPL模型的优化拟合。该数据集主要用于评估和改进3D人体姿态估计方法,特别是在自然环境中的应用,如虚拟/增强现实和机器人技术。
The SMPL Mannequin Benchmark is a dataset containing 24,428 image frames, developed by NAVER LABS Europe for evaluating 3D human pose estimation in natural environments. It leverages static human videos from the Mannequin Challenge dataset, and accurately fits the SMPL model via structure-from-motion (SfM) techniques. The dataset covers diverse body poses, visual appearances and environmental scenarios, including both indoor and outdoor scenes, natural occlusions and close-up shots. Its creation involves human instance tracking and segmentation, 3D point cloud reconstruction, and optimized fitting of the SMPL model. This benchmark is primarily used to evaluate and improve 3D human pose estimation methods, especially for real-world applications such as virtual/augmented reality and robotics.

- 1SMPLy Benchmarking 3D Human Pose Estimation in the WildNAVER LABS Europe · 2020年



