Cross-Model Distillation of a Human-Pose Foundation Model from Unannotated Infant Video for Markerless 3D Pose Estimation
研究通过从未标注的婴儿视频中进行跨模型蒸馏,提高3D姿态估计模型在婴儿上的表现,解决现有模型在婴儿上不准确的问题。
研究通过从未标注的婴儿视频中进行跨模型蒸馏,提高3D姿态估计模型在婴儿上的表现,解决现有模型在婴儿上不准确的问题。
研究通过在SAM-3D-Body模型上增加一个生物力学预测头,利用自监督蒸馏方法从单张RGB图像中回归出生物力学模型的关节角度和尺度,解决了现有方法输出缺乏生物力学定义的关节角度的问题。
Clinical practice lacks accessible, objective tools for motor function assessment, hindering the adoption of biomechanical metrics in rehabilitation and neurology. To address this, we propose a portable, smartphone-based monocular video analysis framework: secure video acquisition via mobile devices is coupled with cloud-based computation, high-precision monocular pose estimation, subject-specific biomechanical modeling, and optimization algorithms—enabling clinical-grade full-body kinematic measurement (joint angle error <3°). Our method significantly outperforms patient-reported outcomes (e.g., mJOA), demonstrating high sensitivity and responsiveness to surgical intervention. It reliably quantifies gait parameters and sensitively detects pre- to postoperative functional changes in cervical spondylotic myelopathy. This work represents the first smartphone-video-driven, clinically validated objective assessment of motor function, establishing a standardized pathway for integrating mobile health technologies into evidence-based rehabilitation medicine.
研究通过从未标注的婴儿视频中进行跨模型蒸馏,提高3D姿态估计模型在婴儿上的表现,解决现有模型在婴儿上不准确的问题。
研究通过在SAM-3D-Body模型上增加一个生物力学预测头,利用自监督蒸馏方法从单张RGB图像中回归出生物力学模型的关节角度和尺度,解决了现有方法输出缺乏生物力学定义的关节角度的问题。
Clinical practice lacks accessible, objective tools for motor function assessment, hindering the adoption of biomechanical metrics in rehabilitation and neurology. To address this, we propose a portable, smartphone-based monocular video analysis framework: secure video acquisition via mobile devices is coupled with cloud-based computation, high-precision monocular pose estimation, subject-specific biomechanical modeling, and optimization algorithms—enabling clinical-grade full-body kinematic measurement (joint angle error <3°). Our method significantly outperforms patient-reported outcomes (e.g., mJOA), demonstrating high sensitivity and responsiveness to surgical intervention. It reliably quantifies gait parameters and sensitively detects pre- to postoperative functional changes in cervical spondylotic myelopathy. This work represents the first smartphone-video-driven, clinically validated objective assessment of motor function, establishing a standardized pathway for integrating mobile health technologies into evidence-based rehabilitation medicine.