Part-Level Graph Convolutional Network for Skeleton-Based Action Recognition

Authors: Linjiang Huang, Yan Huang, Wanli Ouyang, Liang Wang11045-11052

AAAI 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Extensive experiments on two benchmark datasets show the state-of-the-art performance on skeleton-based action recognition and demonstrate the effectiveness of the proposed method.
Researcher Affiliation Collaboration Linjiang Huang,1,3 Yan Huang,1,3 Wanli Ouyang,4 Liang Wang1,2,3 1Center for Research on Intelligent Perception and Computing (CRIPAC) National Laboratory of Pattern Recognition (NLPR) 2Center for Excellence in Brain Science and Intelligence Technology (CEBSIT) Institute of Automation, Chinese Academy of Sciences (CASIA) 3University of Chinese Academy of Sciences (UCAS) 4University of Sydney Sense Time Computer Vision Research Group, Australia
Pseudocode No The paper does not contain any structured pseudocode or algorithm blocks labeled as 'Algorithm' or 'Pseudocode'.
Open Source Code No The paper does not provide an explicit statement about releasing source code for the described methodology or a link to a code repository.
Open Datasets Yes We evaluate our method on two challenging datasets, namely NTU RGB+D (Shahroudy et al. 2016) and SYSU 3D HOI (SYSU) (Hu et al. 2015).
Dataset Splits Yes NTU RGB+D. This dataset consists of 56880 actions with 60 classes. The benchmark evaluations include Cross Subject (CS) and Cross-View (CV). In the CS evaluation, training samples come from one subset of actors and networks are evaluated on samples from remaining actors. In the CV evaluation, samples captured from cameras 2 and 3 are utilized for training, while samples from camera 1 are employed for testing. SYSU. This dataset contains 12 actions performed by 40 subjects. We follow the protocols proposed by (Hu et al. 2015) to evaluate the performance. For setting 1, for each activity class, half of the samples are used for training and the rest for testing. For setting 2, half of the subjects are used for training and the rest for testing. For each setting, we employ the 30-fold cross-subject validation.
Hardware Specification No The paper mentions 'NVIDIA DGX-1 AI Supercomputer' in the acknowledgments, but does not explicitly state that this hardware was used for running the experiments or provide specific details like GPU or CPU models, or memory.
Software Dependencies No The proposed model is implemented by Pytorch (Paszke et al. 2017).
Experiment Setup Yes We set the number of body parts as 6 for the two blocks. The proposed model is implemented by Pytorch (Paszke et al. 2017). We use SGD to optimize the model with a mini-batch size of 64. The learning rates for both datasets are 0.1 initially, multiplied by 0.1 after 20 epochs and 50 epochs. The training procedure stops at 80 epochs.