STIT: Spatio-Temporal Interaction Transformers for Human-Object Interaction Recognition in Videos

Almushyti, Muna; Li, Frederick W.B.

Abstract

Recognizing human-object interactions is challenging due to their spatio-temporal changes. We propose the SpatioTemporal Interaction Transformer-based (STIT) network to reason such changes. Specifically, spatial transformers learn humans and objects context at specific frame time. Temporal transformer then learns the relations at a higher level between spatial context representations at different time steps, capturing longterm dependencies across frames. We further investigate multiple hierarchy designs in learning human interactions. We achieved superior performance on Charades, Something-Something v1 and CAD-120 datasets, comparing to baseline models without learning human-object relations, or with prior graph-based networks. We also achieved state-of-the-art accuracy of 95.93% on CAD-120 dataset [1] by employing RGB data only.

Citation

Almushyti, M., & Li, F. W. (2022, August). STIT: Spatio-Temporal Interaction Transformers for Human-Object Interaction Recognition in Videos. Presented at 2022 26th International Conference on Pattern Recognition (ICPR), Montréal, Québec

• Presentation Conference Type: Conference Paper (published)
• Conference Name: 2022 26th International Conference on Pattern Recognition (ICPR)
• Start Date: Aug 21, 2022
• End Date: Aug 25, 2022
• Acceptance Date: May 17, 2022
• Publication Date: 2022-11
• Deposit Date: Oct 31, 2022
• Publicly Available Date: Nov 1, 2022
• Publisher: Institute of Electrical and Electronics Engineers
• Pages: 3287-3294
• DOI: https://doi.org/10.1109/icpr56361.2022.9956030
• Public URL: https://durham-repository.worktribe.com/output/1135752
• Related Public URLs: https://doi.org/10.1109/ICPR56361.2022.9956030
• Additional Information: 21-25 Aug. 2022