AcinoSet

AcinoSet是野外自由运行的猎豹的数据集，其中包含119,490帧的多视图同步高速视频素材，相机校准文件和7,588的人类注释帧。我们利用无标记动物姿势估计与DeepLabCut提供2D关键点 (在119K帧中)。然后，我们使用三种方法作为3D姿势估计工具开发的强基线: 传统的稀疏束调整，扩展的卡尔曼滤波器和基于轨迹优化的方法，我们称为全轨迹估计。还提供了生成的3D轨迹，人工检查的3D地面真相以及用于检查数据的交互式工具。我们相信该数据集将对生态学，机器人学，生物力学以及计算机视觉等各种领域有用。

AcinoSet is a dataset of freely running cheetahs in the wild, which contains synchronized multi-view high-speed video footage with 119,490 frames, camera calibration files, and 7,588 manually annotated frames. We employed markerless animal pose estimation via DeepLabCut to generate 2D keypoints for all 119K frames. Subsequently, we developed three strong baselines for 3D pose estimation: traditional sparse bundle adjustment, extended Kalman filter, and a trajectory optimization-based method we refer to as full trajectory estimation. Additionally, the generated 3D trajectories, manually verified 3D ground truth, and interactive tools for data inspection are provided. We believe this dataset will be valuable for a wide range of fields including ecology, robotics, biomechanics, and computer vision.

搜集汇总

背景与挑战

背景概述

AcinoSet是一个包含119,490帧多视图同步高速视频的猎豹数据集，提供7,588帧人类注释和2D/3D关键点估计，适用于生态学、机器人学、生物力学和计算机视觉等领域的研究。