Tag
Maniformer 的新论文提出 MEgoVista，一套基于头戴式装置的物理 AI 数据系统，通过立体视觉和全序列优化，将手部手指误差降至 4.29 毫米、头部跟踪误差控制在 0.65–2.83 毫米，并经 26 相机动作捕捉实验室验证，为机器人学习提供可测量的人类动作数据。
JoHan：一个从视频序列直接联合生成2D和3D手部姿态序列的统一生成框架，无需依赖逐帧姿态预测，在遮挡和不完整观测下显著提升手部运动重建的精度、速度和平滑度。
MEgoVista is an offline pipeline from Maniformer that turns unprepared egocentric MEgo View recordings into metric two-hand and head motion in a gravity-aligned world frame, using calibrated stereo for metric scale and validating outputs against independent Chingmu optical motion capture. The work positions itself as a scalable, unconstrained alternative to studio rigs for generating metric hand supervision for robot manipulation learning.
EasyMocap is an open-source toolbox built by Zhejiang University that enables 3D human pose estimation from regular RGB videos, used by research projects like HumanNeRF, but with licensing considerations due to SMPL body models.
A fast and affordable alternative to traditional motion capture, featuring Sara Silkin.
Proposes physical self-supervised learning, an autoencoder paradigm for label-free IMU sensing that replaces the neural decoder with a physics-based decoder, achieving up to 5x error reduction in tracking and motion capture tasks without manual labels.
Sara Silkin presents a new AI motion capture pipeline, likely building upon the techniques used in 'Jibaro'.
VFX artist demonstrates how to use Uisato Studio's Kling Motion Control to transform an iPhone dance video into a multi-angle robotic aesthetic piece, showcasing low-budget motion capture techniques.
Explains that 'markerless' motion capture still uses markers but estimates them via webcam instead of expensive equipment, enabling multi-person, scale-aware capture.
The article announces the release of Marionette, a browser-based tool that allows users to record and share robot motions for the Reachy Mini using their phones or computers.
MoCapAnything V2 introduces a fully end-to-end framework for arbitrary-skeleton motion capture from monocular video, jointly optimizing video-to-pose and pose-to-rotation predictions to resolve rotation ambiguity.
A 20-year-old U.S. student built millimeter-accurate motion-capture hardware with $3 in parts and Claude Code overnight, then sold the project for $230k a week later.