This is the source code for our paper: Automated Spatial-Temporal Graph Neural Network Search for Skeleton-based Human Action Recognition on Edge Devices. A brief introduction of this work is as follows:
Graph neural networks (GNNs) have attracted significant attention in human action recognition (HAR) tasks due to their ability to model spatial-temporal relationships between body joints in skeletal graphs. However, prior solutions typically suffer from high computational overhead and significant inference latency, particularly on resource-constrained platforms like edge devices. To mitigate this, we propose Edge-STGNN, a speed-optimized neural architecture search (NAS) framework, for the automated design of spatial-temporal GNNs (STGNNs) tailored for action recognition on edge devices. Specifically, Edge-STGNN defines a search space that includes diverse temporal convolutional layers, attention mechanisms, and network depth options. Based on this search space, it constructs the supernet using a single-path training approach. Subsequently, an evolutionary algorithm-based search strategy is applied to identify optimal architectures. Furthermore, Edge-STGNN integrates an efficient speed predictor to reduce evaluation time, and employs a small-scale proxy dataset to lower search costs. Comprehensive experiments on multiple datasets and edge devices demonstrate that Edge-STGNN effectively identifies architectures with fewer parameters, reduced computational complexity, and faster inference speeds, while preserving acceptable recognition accuracy. Additionally, by incorporating speed constraints, Edge-STGNN enables a flexible trade-off between accuracy and speed tailored to specific application demands. For example, on the NTU RGB+D 60 xview60 benchmark, compared to all baseline methods with reported speeds, Edge-STGNN achieves a 2.4$\times$ to 9.4$\times$ inference speedup while maintaining accuracy within a ±1% margin.
图神经网络（GNN）因能够建模骨骼图中关节点之间的时空关系，在人体动作识别（HAR）任务中引起了广泛关注。然而，现有方法通常存在计算开销高、推理延迟大的问题，尤其是在边缘设备等资源受限平台上。为解决这一问题，我们提出了 Edge-STGNN，一个面向速度优化的神经架构搜索（NAS）框架，用于自动化设计适合边缘设备上动作识别的时空图神经网络（STGNN）。具体而言，Edge-STGNN 定义了一个搜索空间，包含多种时间卷积层、注意力机制和网络深度选项。基于该搜索空间，它采用单路径训练方式构建超网络。随后，应用基于进化算法的搜索策略来识别最优架构。此外，Edge-STGNN 集成了一个高效的速度预测器以减少评估时间，并采用小规模代理数据集来降低搜索成本。在多个数据集和边缘设备上的全面实验表明，Edge-STGNN 能够有效识别出参数更少、计算复杂度更低、推理速度更快且保持可接受识别精度的架构。通过引入速度约束，Edge-STGNN 还能根据具体应用需求在精度与速度之间实现灵活权衡。例如，在 NTU RGB+D 60 xview60 基准上，与所有报告了速度的基线方法相比，Edge-STGNN 在精度保持 ±1% 范围内的同时，实现了 2.4 倍到 9.4 倍的推理加速。
@ARTICLE{11516157,
author={Li, Xiuwen and Fang, Weiwei and Yue, Liang and Shi, Fanjie and Xiong, Neal N.},
journal={IEEE Internet of Things Journal},
title={Automated Spatial-Temporal Graph Neural Network Search for Skeleton-based Human Action Recognition on Edge Devices},
year={2026},
volume={},
number={},
pages={1-1},
keywords={Modeling;Accuracy;Computer architecture;Architecture;Skeleton;Human activity recognition;Design methodology;Neural architecture search;Hardware;Internet of Things;Graph neural networks;neural architecture search;human activity recognition;edge computing},
doi={10.1109/JIOT.2026.3692679}
}
This work will be published by IEEE Internet of Things Journal. Click here for our paper. Source code and data will be made available on request.