Scholarlink

<p>We present a wearable IMU–based human pose estimation framework that couples knowledge distillation with an involution–based student and a principled structural re–parameterization for on–device deployment. A high–capacity Transformer serves as the teacher to learn rich temporal–spatial representations, while the student adopts input–adaptive involution operators. During deployment, structural re–parameterization collapses the training graph by folding batch normalization and fusing in–branch cascades and cross–branch parallel paths, yielding a single inference–time module that is equivalent to two 1D CNN passes. This design decouples training expressiveness from inference efficiency and makes the model hardware–friendly for low–power wearables. Extensive experiments on two public benchmarks, DIP–IMU and IMUPoser, demonstrate that our approach preserves near state–of–the–art accuracy while achieving sub–millisecond latency. Concretely, the proposed model attains 81&#xa0;mm MPJPE on DIP–IMU and 94&#xa0;mm on IMUPoser, with per–frame latencies of 0.012 ms and 0.011 ms, respectively—delivering one to two orders of magnitude speedups over heavy Transformer baselines and matching the best accuracies within ≤ 1.25% relative difference. The consistent gains across datasets indicate strong robustness and cross–subject generalization, highlighting the suitability of the method for real–time wearable applications.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Human-centered design-based lightweight wearable IMU human pose estimation

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年2月28日

查看全文
  • Lidong Wang,
  • Juanjuan Liu,
  • Jingxuan Xue,
  • Qinyu Tan,
  • Minmin Li

摘要

We present a wearable IMU–based human pose estimation framework that couples knowledge distillation with an involution–based student and a principled structural re–parameterization for on–device deployment. A high–capacity Transformer serves as the teacher to learn rich temporal–spatial representations, while the student adopts input–adaptive involution operators. During deployment, structural re–parameterization collapses the training graph by folding batch normalization and fusing in–branch cascades and cross–branch parallel paths, yielding a single inference–time module that is equivalent to two 1D CNN passes. This design decouples training expressiveness from inference efficiency and makes the model hardware–friendly for low–power wearables. Extensive experiments on two public benchmarks, DIP–IMU and IMUPoser, demonstrate that our approach preserves near state–of–the–art accuracy while achieving sub–millisecond latency. Concretely, the proposed model attains 81 mm MPJPE on DIP–IMU and 94 mm on IMUPoser, with per–frame latencies of 0.012 ms and 0.011 ms, respectively—delivering one to two orders of magnitude speedups over heavy Transformer baselines and matching the best accuracies within ≤ 1.25% relative difference. The consistent gains across datasets indicate strong robustness and cross–subject generalization, highlighting the suitability of the method for real–time wearable applications.