Live and Learn: Continual Action Clustering with Incremental Views

Authors: Xiaoqiang Yan, Yingtao Gan, Yiqiao Mao, Yangdong Ye, Hui Yu

AAAI 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental The empirical experimental results on 6 realistic multi-view action collections demonstrate the excellent clustering performance and time/space efficiency of the CAC compared with 15 state-of-the-art baselines.
Researcher Affiliation Academia Xiaoqiang Yan1, Yingtao Gan1, Yiqiao Mao1, Yangdong Ye1 , Hui Yu2 1 School of Computer and Artificial Intelligence, Zhengzhou University, Zhengzhou, China 2 School of Creative Technologies, University of Portsmouth, PO1 2DJ, United Kingdom
Pseudocode Yes Algorithm 1: Continual Action Clustering (CAC) Algorithm
Open Source Code No The paper states, 'We perform all baselines according to the original parameter settings from their papers, and the source codes are available from original authors or websites.' This refers to baseline code, not the code for the proposed CAC method, and no direct link or explicit statement of release for their own code is provided.
Open Datasets Yes We adopt the following widely-used multi-view human action datasets to evaluate the effectiveness of the CAC. They are: IXMAS (Ramagiri, Kavi, and Kulathumani 2011), MMI (Liu et al. 2017), MSR (Xu et al. 2016), WVU (Weinland, Boyer, and Ronfard 2007), UCLA (Wang et al. 2014), MV-TJU (Liu et al. 2015)... To facilitate easier comparison for readers, we apply the CAC to more datasets widely used in the MVC baselines, i.e., CCV (Jiang et al. 2011), Cite Seer1 and You Tube (Madani, Georg, and Ross 2013). 1https://linqs-data.soe.ucsc.edu/public/lbc/citeseer.tgz
Dataset Splits No The paper does not provide specific details on dataset splits (e.g., percentages or sample counts for training, validation, and testing sets), nor does it explicitly mention a validation set.
Hardware Specification Yes All experiments are performed on a desktop computer with RTX 4090 GPU, i7-13700K CPU, 64G RAM, Windows 10 system.
Software Dependencies No The paper mentions 'We implement the deep MVC baselines on Py Torch toolbox. Our proposed CAC and other multi-view clustering baselines are conducted on Matlab 2022a.' While Matlab 2022a has a version, PyTorch does not, and no specific versions for other libraries or dependencies are provided, which is insufficient for full reproducibility of the software environment.
Experiment Setup Yes In our model, there is a regularization parameter λ that balances the weights between historical and new coming views. Similar with the late fusion MVC baselines (CMVC, LFMVC, LKA), we tune λ in the range of 2. [ 10, 9, , 9, 10].