Offensive Play Recognition of Basketball Video Footage Using ActionFormer

Tafadzwa Blessing Chiura, Dustin van der Haar

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review


This paper will aim to conduct 3 experiments to determine the best-performing action recognition approach on the SpaceJam dataset. The 3 experiments are Temporal Segment Network (TSN), Inflated 3D-CNN (I3D) and Pose-estimation (Pose-C3D). TSN and I3D yielded similar results with TSN scoring 54.88% for the mean accuracy, 94.33% for top-5 accuracy and 54.88 top-1 accuracy. And I3D scored 53.07% mean accuracy, 91.65% for top-5 accuracy and 53.07 mean accuracy. When Pose-C3d is run for 240 epochs it achieves better results with a top 1 accuracy and mean-class accuracy of 63.15% and a top-5 accuracy of 95.51. These results indicate that the models can distinguish between similar actions such as running and walking in basketball with relative success.

Original languageEnglish
Title of host publicationHCI International 2023 Posters - 25th International Conference on Human-Computer Interaction, HCII 2023, Proceedings, Part I
EditorsConstantine Stephanidis, Margherita Antona, Stavroula Ntoa, Gavriel Salvendy
PublisherSpringer Science and Business Media Deutschland GmbH
Number of pages8
ISBN (Print)9783031359880
Publication statusPublished - 2023
Event25th International Conference on Human-Computer Interaction , HCII 2023 - Copenhagen, Denmark
Duration: 23 Jul 202328 Jul 2023

Publication series

NameCommunications in Computer and Information Science
Volume1832 CCIS
ISSN (Print)1865-0929
ISSN (Electronic)1865-0937


Conference25th International Conference on Human-Computer Interaction , HCII 2023


  • Action Recognition
  • Basketball
  • Offensive Play Recognition

ASJC Scopus subject areas

  • General Computer Science
  • General Mathematics


Dive into the research topics of 'Offensive Play Recognition of Basketball Video Footage Using ActionFormer'. Together they form a unique fingerprint.

Cite this