Research map / Time series / Long-term forecasting
Space-Time-Separable Graph Convolutional Network for Pose Forecasting
Theodoros Sofianos, Alessio Sampieri, Luca Franco, Fabio Galasso
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 |
|---|---|---|---|
| factor the space-time adjacency into separate spatial and temporal matrices within a single graph (vs separate GCNs for space and time) (ablation) | STS-GCN · Human3.6M average, 80 ms | MPJPE (mm): 28.9 -> 10.1 | Improved |
| factor the space-time adjacency into separate spatial and temporal matrices within a single graph (vs separate GCNs for space and time) (ablation) | STS-GCN · Human3.6M average, 1000 ms | MPJPE (mm): 79.9 -> 75.6 | Improved |
| factor the adjacency into A^s A^t instead of a full space-time matrix (ablation) | STS-GCN · Human3.6M average, 80 ms | MPJPE (mm): 11.9 -> 10.1 | Improved |
| factor the adjacency into A^s A^t instead of a full space-time matrix (ablation) | STS-GCN · Human3.6M average, 1000 ms | MPJPE (mm): 82.5 -> 75.6 | Improved |
| use layer-specific rather than shared adjacency matrices (ablation) | STS-GCN · Human3.6M average, 80 ms | MPJPE (mm): 11.3 -> 10.1 | Improved |
| use layer-specific rather than shared adjacency matrices (ablation) | STS-GCN · Human3.6M average, 1000 ms | MPJPE (mm): 76.9 -> 75.6 | Improved |
| use layer-specific rather than shared adjacency matrices (parameter cost) (ablation) | STS-GCN · Human3.6M parameters | parameters: 36.4k -> 57.5k | Hurt |
| use STS-GCN instead of DCT-RNN-GCN | STS-GCN · Human3.6M average, 1000 ms | MPJPE (mm): 112.1 -> 75.6 | Improved |
| use STS-GCN instead of DCT-RNN-GCN | STS-GCN · Human3.6M average, 80 ms | MPJPE (mm): 10.4 -> 10.1 | Improved |
| use STS-GCN instead of DCT-RNN-GCN | STS-GCN · 3DPW (trained on AMASS) | MPJPE (mm): 73.7 -> 42.3 | Improved |
| use STS-GCN instead of DCT-RNN-GCN | STS-GCN · AMASS-BMLrub, 1000 ms | MPJPE (mm): 67.2 -> 45.5 | Improved |
| use STS-GCN instead of DCT-RNN-GCN in parameters | STS-GCN · Human3.6M parameters | parameters: 3.4M -> 57.5k | Improved |
Closest papers by meaning
- Learning Trajectory Dependencies for Human Motion Prediction (ICCV 2019 · similarity 0.92)
- MSR-GCN: Multi-Scale Residual Graph Convolution Networks for Human Motion Prediction (ICCV 2021 · similarity 0.91)
- Progressively Generating Better Initial Guesses Towards Next Stages for High-Quality Human Motion Prediction (CVPR 2022 · similarity 0.91)
- Learning Dynamics via Graph Neural Networks for Human Pose Estimation and Tracking (CVPR 2021 · similarity 0.90)
- Modulated Graph Convolutional Network for 3D Human Pose Estimation (ICCV 2021 · similarity 0.89)
- GCNext: Towards the Unity of Graph Convolutions for Human Motion Prediction (AAAI 2024 · similarity 0.89)
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.
