Liu, Jiang, Huang, Qiqi, Li, Yixiao, Ma, Yueran, Fu, Yao, Liu, Xiaochang, Wu, Yingying, Zhou, Wei, Stawarz, Katarzyna  ORCID: https://orcid.org/0000-0001-9021-0615 and Liu, Hantao ORCID: https://orcid.org/0000-0003-4544-3481
2026.
Saliency-guided action quality assessment: an AI-augmented framework for skill evaluation in physical education.
Presented at: 2026 IEEE Conference on Artificial Intelligence (CAI),
Granada, Spain,
08 - 10 May 2026.
2026 IEEE Conference on Artificial Intelligence (CAI) Proceedings.
2026 IEEE Conference on Artificial Intelligence (CAI).
IEEE,
pp. 1997-2002.
10.1109/cai68641.2026.11536588
|
Abstract
Action Quality Assessment (AQA) is increasingly vital for providing objective, scalable feedback in higher education physical training. However, most current AQA methods rely on global features and fail to mimic the selective attention mechanism that human experts use to diagnose student performance. In this paper, we hypothesize that incorporating visual saliency can enhance AQA by capturing critical spatiotemporal information in action sequences. To this end, we propose a fusion framework - Saliency-Guided Action Quality Assessment Network (SAQANet) - that integrates saliency features into AQA models. We evaluate the impact of saliency integration by applying six state-of-the-art video saliency prediction models within the SAQANet framework. Experimental results demonstrate that saliency-guided AQA models can achieve significant performance improvements, offering a pathway towards more transparent and reliable AI teaching assistants.
| Item Type: | Conference or Workshop Item - published (Paper) |
|---|---|
| Date Type: | Published Online |
| Status: | Published |
| Schools: | Schools > Computer Science & Informatics |
| Publisher: | IEEE |
| ISBN: | 9798331560393 |
| Last Modified: | 25 Aug 2026 21:47 |
| URI: | https://orca.cardiff.ac.uk/id/eprint/187615 |
Actions (repository staff only)
 |
Edit Item |
Dimensions
Dimensions