Summary of Rel-sar: Representation Learning For Skeleton Action Recognition with Convolutional Transformers and Byol, by Safwen Naimi et al.
ReL-SAR: Representation Learning for Skeleton Action Recognition with Convolutional Transformers and BYOL
by Safwen Naimi, Wassim Bouachir, Guillaume-Alexandre Bilodeau
First submitted to arxiv on: 9 Sep 2024
Categories
- Main: Computer Vision and Pattern Recognition (cs.CV)
- Secondary: Artificial Intelligence (cs.AI)
GrooveSquid.com Paper Summaries
GrooveSquid.com’s goal is to make artificial intelligence research accessible by summarizing AI papers in simpler terms. Each summary below covers the same AI paper, written at different levels of difficulty. The medium difficulty and low difficulty versions are original summaries written by GrooveSquid.com, while the high difficulty version is the paper’s original abstract. Feel free to learn from the version that suits you best!
| Summary difficulty | Written by | Summary |
|---|---|---|
| High | Paper authors | High Difficulty Summary Read the original abstract here |
| Medium | GrooveSquid.com (original content) | Medium Difficulty Summary The paper presents a novel unsupervised representation learning framework for skeleton action recognition, dubbed ReL-SAR. This approach leverages the combination of convolutional and attention layers to model spatial and temporal cues in skeleton sequences. The authors also introduce a Selection-Permutation strategy for skeleton joints to enhance informative descriptions from skeletal data. To learn robust representations from unlabeled sequence data, they employ Bootstrap Your Own Latent (BYOL). The method is evaluated on four limited-size datasets: MCAD, IXMAS, JHMDB, and NW-UCLA, achieving competitive results against state-of-the-art methods in terms of both performance and computational efficiency. |
| Low | GrooveSquid.com (original content) | Low Difficulty Summary This paper creates a way to recognize actions from skeleton data without needing lots of labeled information. It uses special kinds of neural networks called convolutional transformers to learn features that work well for different action recognition tasks. The approach also includes techniques to select the most important parts of the skeleton data and use unlabeled data to learn better representations. The method is tested on several small datasets and shows good results compared to other approaches. |
Keywords
» Artificial intelligence » Attention » Representation learning » Unsupervised
