Unsupervised Motion Representation Learning with Capsule Autoencoders
Authors: Ziwei Xu, Xudong Shen, Yongkang Wong, Mohan S. Kankanhalli
NeurIPS 2021 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | MCAE is evaluated on a novel Trajectory20 motion dataset and various real-world skeleton-based human action datasets. Notably, it achieves better results than baselines on Trajectory20 with considerably fewer parameters and state-of-the-art performance on the unsupervised skeleton-based action recognition task. |
| Researcher Affiliation | Academia | Ziwei Xu , Xudong Shen , Yongkang Wong , Mohan S Kankanhalli School of Computing, National University of Singapore NUS Graduate School, National University of Singapore {ziwei-xu, mohan}@comp.nus.edu.sg xudong.shen@u.nus.edu, yongkang.wong@nus.edu.sg |
| Pseudocode | No | The paper describes the methodology in prose and mathematical equations but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | The source code and the T20 dataset of our research are accessible at https://github.com/Ziwei XU/Capsule Motion. |
| Open Datasets | Yes | We propose Trajectory20, a novel and challenging synthetic dataset with a wide class of motion patterns and controllable intra-class variations. ... The source code and the T20 dataset of our research are accessible at https://github.com/Ziwei XU/Capsule Motion. ... Three widely-used datasets are used for evaluation: NW-UCLA [48], NTU-RGBD60 (NTU60) [37], and NTU-RGBD120 (NTU120) [25]. |
| Dataset Splits | Yes | The training data is generated on-the-fly and a fixed test set of 10,000 samples is used for evaluation. ... For NTU60, we follow the official data split for the cross-subject (XSUB) and cross-view (XVIEW) protocols. The similar is implemented on NTU120 for the cross-subject (XSUB) and cross-setting (XSET) protocol. |
| Hardware Specification | Yes | The experiments are run on an NVIDIA Titan V GPU, where we use a batch size of 64, and the Adam [16] optimizer with a learning rate of 10 3. ... The computational work for this article was partially performed on resources of the National Supercomputing Centre, Singapore (https://www.nscc.sg). |
| Software Dependencies | No | The paper mentions using the Adam optimizer but does not specify version numbers for any programming languages, libraries, or other software dependencies. |
| Experiment Setup | Yes | We report the mean accuracy and standard error based on three runs with random initialization. The experiments are run on an NVIDIA Titan V GPU, where we use a batch size of 64, and the Adam [16] optimizer with a learning rate of 10 3. Please refer to the supplementary material for details. |