A Generative Model for Recognizing Mixed Group Activities in Still Images

Authors: Zheng Zhou, Kan Li, Xiangjian He, Mengmeng Li

IJCAI 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experimental results demonstrate that our model makes good interpretations for mixed group activities and outperforms the state-of-the-art methods on the Collective Activity Classification dataset. To compare with other approaches, we conduct experiments on the popular Collective Activity Classification (CAC) dataset [Choi et al., 2009], on which many state-of-the-art methods have been performed. The experimental results demonstrate that our model produces good interpretations for mixed group activities and outperforms all of the state-of-the-art methods on this dataset.
Researcher Affiliation Academia 1School of Computer, Beijing Institute of Technology, Beijing, China 2School of Computing and Communications, University of Technology Sydney, Sydney, Australia
Pseudocode No The paper describes the generation process and mathematical distributions but does not include any structured pseudocode or algorithm blocks.
Open Source Code No The paper provides a link for a dataset ('Dataset: http://cs.bit.edu.cn/szdw/jsml/js/lk/index.htm. 2016.') but does not explicitly state that the source code for the methodology described in the paper is being released or provide a link to it.
Open Datasets Yes To compare with other approaches, we select the popular Collective Activity Classification (CAC) dataset, on which many excellent approaches have been tested [Choi et al., 2009]. A new set of labels that annotate the positions of hu-man s body parts on all images in the CAC dataset is produced and used as the ground truths of human poses. Furthermore, a Special dataset containing all of the CAC images which contain multiple group activities is also created. These two datasets are released for others to use when working on the same areas for recognition of mixed group activities. [Zhou et al., 2016] Zheng Zhou, Kan Li, Xiangjian He, and Li Mengmeng. Dataset: http://cs.bit.edu.cn/szdw/jsml/js/lk/index.htm. 2016.
Dataset Splits Yes We apply 5-fold cross validation to evaluate our model on the formed dataset.
Hardware Specification No The paper does not provide specific hardware details such as CPU, GPU models, or memory specifications used for running its experiments.
Software Dependencies No The paper mentions using a 'multi-scale deformable part based model' and 'SVMs' but does not provide specific version numbers for any software dependencies, libraries, or programming languages used in the implementation or experiments.
Experiment Setup Yes In these experiments, we learn 6 SVMs of standard poses for each activity and set the ratio of : β to be 1. We also show the experimental results about the effects of the number of SVMs for each activity and the ratio of : β.