Ske2Grid: Skeleton-to-Grid Representation Learning for Action Recognition

Authors: Dongqi Cai, Yangyuxuan Kang, Anbang Yao, Yurong Chen

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We construct networks upon prevailing graph convolution networks and conduct experiments on six mainstream skeleton-based action recognition datasets. Experiments show that our Ske2Grid significantly outperforms existing GCN-based solutions under different benchmark settings, without bells and whistles. Code and models are available at https: //github.com/OSVAI/Ske2Grid.
Researcher Affiliation Industry 1Intel Labs China. Correspondence to: Anbang Yao <anbang.yao@intel.com>.
Pseudocode No The paper describes methods like GIT, UPT, and PLS in text but does not provide structured pseudocode or algorithm blocks.
Open Source Code Yes Code and models are available at https: //github.com/OSVAI/Ske2Grid.
Open Datasets Yes NTU-60 (Shahroudy et al., 2016) is the first large-scale multi-modality skeleton-based action recognition dataset.
Dataset Splits Yes There are two popular validation protocols for this dataset: cross-subject (XSub) and cross-view (XView).
Hardware Specification No The paper does not provide specific hardware details (like CPU/GPU models, memory) used for running the experiments.
Software Dependencies No The paper mentions software like PYSKL, HRNet, Faster-RCNN, and ResNet50, but it does not specify version numbers for these or other key software dependencies required for replication.
Experiment Setup Yes For the experiments conducted to explore the effects of our core designs, we use the vanilla training strategy of PYSKL as in ST-GCN (Yan et al., 2018) for fair and clean comparisons, in which each model is trained for 80 epochs with the learning rate decayed by 10 at 10th and 50th epochs respectively. For the main experiments to explore the capability of Ske2Grid as shown in Table 2,Table 3, Table 7 and Table 8, we use the latest common experimental setups in PYSKL, in which each model is trained for 80 epochs with the cosine schedule of learning rate. In both settings, the initial learning rate is set to 0.1, the batch size is 128, the momentum is set to 0.9, the weight decay is 5 10 4, and the Nesterov momentum is used for the optimizer.