Research map / Learning theory and algorithms / Matrix and tensor methods
GeomNet: A Neural Network Based on Riemannian Geometries of SPD Matrix Space and Cholesky Space for 3D Skeleton-Based Interaction Recognition
Xuan Son Nguyen
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add parallel transport in covariance computation (ablation) | GeomNet · SBU Interaction | accuracy (%): 71.51 -> 96.33 | Improved |
| add parallel transport in covariance computation (ablation) | GeomNet · NTU RGB+D 60 X-Subject | accuracy (%): 62.18 -> 93.62 | Improved |
| add parallel transport in covariance computation (ablation) | GeomNet · NTU RGB+D 60 X-View | accuracy (%): 66.83 -> 96.32 | Improved |
| add lower triangular matrix learning layer (ablation) | GeomNet · SBU Interaction | accuracy (%): 94.90 -> 96.33 | Improved |
| add lower triangular matrix learning layer (ablation) | GeomNet · NTU RGB+D 60 X-Subject | accuracy (%): 92.30 -> 93.62 | Improved |
| add lower triangular matrix learning layer (ablation) | GeomNet · NTU RGB+D 60 X-View | accuracy (%): 95.05 -> 96.32 | Improved |
| use (k,k')=(2,3) instead of (1,1) in Gaussian embedding (ablation) | GeomNet · SBU Interaction | accuracy (%): 94.54 -> 96.33 | Improved |
| embed Gaussians with k=10 instead of k=1 in SPDNet (ablation) | SPDNet · SBU Interaction | accuracy (%): 90.5 -> 92.38 | Improved |
| add the mean vector via Gaussian embedding (k=10) rather than covariance only (k=0) (ablation) | SPDNet · SBU Interaction | accuracy (%): 79.48 -> 92.38 | Improved |
| use GeomNet instead of SPDNet | SPDNet · SBU Interaction | accuracy (%): 92.38 -> 96.33 | Improved |
| use GeomNet instead of ST-GCN-PAM | ST-GCN-PAM · NTU RGB+D 120 X-Subject | accuracy (%): 83.28 -> 86.49 | Improved |
| use AS-GCN actional and structural links instead of ST-GCN | ST-GCN · NTU RGB+D 60 X-Subject | accuracy (%): 86.75 -> 87.08 | No clear effect |
Later papers that compared with this method
1 later accepted paper reports GeomNet as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- A Gyrovector Space Approach for Symmetric Positive Semi-Definite Matrix Learning (ECCV 2022 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- A Novel Geometric Framework on Gram Matrix Trajectories for Human Behavior Understanding (TPAMI 2018 · similarity 0.87)
- Building Neural Networks on Matrix Manifolds: A Gyrovector Space Approach (ICML 2023 · similarity 0.86)
- A Gyrovector Space Approach for Symmetric Positive Semi-Definite Matrix Learning (ECCV 2022 · similarity 0.85)
- Matrix Manifold Neural Networks++ (ICLR 2024 · similarity 0.85)
- RMLR: Extending Multinomial Logistic Regression into General Geometries (NeurIPS 2024 · similarity 0.83)
- ContactField: Implicit Field Representation for Multi-Person Interaction Geometry (NeurIPS 2024 · similarity 0.83)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
