Reproduce “Logsig-RNN: a novel network for robust and efficient skeleton-based action recognition” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Shujian Liao, Terry Lyons, Weixin Yang, Kevin Schlegel, Hao Ni
The paper introduces LOGSIG-RNN, a network for skeleton-based action recognition that integrates log-signatures from rough path theory with recurrent neural networks. This architecture effectively processes variable-length and irregular skeleton data, achieving state-of-the-art performance on Chalearn2013 while demonstrating enhanced robustness and efficiency compared to standard RNN and GCN baselines.
Key insights
- Log-signatures, derived from rough path theory, offer a mathematically principled way to compactly summarize complex time series data into low-dimensional representations, robust to irregular sampling and invariant to time re-parameterization.
- Integrating a non-trainable, differentiable Log-Signature Layer before an RNN significantly improves efficiency and robustness by providing the RNN with a more compact and informative sequence of local temporal summaries.
- Combining log-signatures with effective spatial feature extraction, such as Embedding Layers or Graph Convolutional Networks, creates a powerful and generic architecture for skeleton-based action recognition.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.