A Spatial-Temporal Graph Convolutional Networks-based Approach for the OpenPack Challenge 2022
A Spatial-Temporal Graph Convolutional Networks-based Approach for the OpenPack Challenge 2022
复制标题
DOI：
10.1109/percomworkshops56833.2023.10150404
复制
发表时间：
2023-03
期刊：
影响因子：
--
通讯作者：
Shurong Chai;Jiaqing Liu;R. Jain;Yinhao Li;T. Tateyama;Yen-Wei Chen
中科院分区：
文献类型：
--
作者：
Shurong Chai;Jiaqing Liu;R. Jain;Yinhao Li;T. Tateyama;Yen-Wei Chen
我们报告了Ritsumei团队为2022年OpenPack挑战赛提出的方法。在这项工作中，我们提出使用运动感知和时间增强的时空图卷积网络来表示关键点模态特征。我们还利用加速度计和陀螺仪模式作为辅助模式来提高性能。我们的最终结果是基于四种模式的融合。我们在提交集上报告了92.32%的F1得分，在2022年OpenPack挑战赛中获得了第三名。
We report the proposed method of Team Ritsumei for the OpenPack challenge 2022. In this work, we proposed to use a motion-aware and temporal-enhanced Spatial-Temporal Graph Convolutional Networks for the representation of the keypoint modality features. We also leverage the Accelerometer and Gyroscope modality as auxiliary modalities to improve the performance. Our final result is based on the fusion of four modalities. We report the 92.32% F1 score on the submission set, which won the 3rd place in the OpenPack challenge 2022.