duality-robotics/pose_estimation5.1

这是一个用于姿态估计的合成数据集，包含1000张图像、掩膜和标签。数据集中的数字双胞胎不是由AI生成的，而是由3D艺术家制作的，与真实世界物体无法区分，这使得基于此数据的训练可以转移到现实世界的应用中。数据集使用FalconEditor模拟软件生成，可以轻松创建数以千计的具有不同光照、姿态、遮挡、背景、摄像机位置等的图像，这有助于模型的稳健训练。标签与数据一起创建，特别是在模拟环境中生成姿态估计标签比在物理世界中容易得多，这不仅节省了大量时间，而且确保了标签的极高准确性和可靠性。

This is a synthetic dataset for pose estimation, containing 1000 images, masks, and labels. The digital twins in the dataset are not generated by AI but crafted by 3D artists to be INDISTINGUISHABLE from physical-world objects, allowing the training from this data to transfer into real-world applicability. The dataset is generated using the FalconEditor simulation software, which can easily create thousands of images with varying lighting, posing, occlusions, backgrounds, camera positions, and more. The labels are created along with the data. The dataset is designed for pose estimation tasks, focusing on the simulation environment.