Hang Wang^1,2 | Zhi-Qi Cheng³ | Youtian Du¹ | Lei Zhang^2
1Xi'an Jiaotong University, ²The Hong Kong Polytechnic University, ³Carnegie Mellon University

We train on the training set of the RepCount-A dataset, and test on the testing set of the RepCount-A dataset, the validation sets of the UCFRep and Countix datasets.

Download datasets: RepCount-A, UCFRep, Countix

on the training set of the RepCount-A dataset

python train.py

on the testing set of the RepCount-A dataset

python test.py

We also provide a pre-trained model for RepCount-A, which can be downloaded from this Google Drive link.

Thanks for works of TransRAC. Our code is based on these implementations.

@ARTICLE{11146674,
  author={Wang, Hang and Cheng, Zhi-Qi and Du, Youtian and Zhang, Lei},
  journal={IEEE Transactions on Multimedia}, 
  title={IVAC-$\mathrm {P^{2}~L}$: Leveraging Irregular Repetition Priors for Improving Video Action Counting}, 
  year={2025},
  volume={},
  number={},
  pages={1-15},
  keywords={Videos;Spatiotemporal phenomena;Visualization;Semantics;Feature extraction;Data mining;Contrastive learning;Training;Complexity theory;Artificial intelligence;Video action counting;irregular repetition priors;inter-cycle consistency;cycle-interval inconsistency},
  doi={10.1109/TMM.2025.3604935}
}

If you have any questions, please feel free to contact: [email protected]