Research map / Graph learning / Knowledge and temporal graphs
STGAT: Modeling Spatial-Temporal Interactions for Human Trajectory Prediction
Yingfan Huang, Huikun Bi, Zhaoxin Li, Tianlu Mao, Zhaoqi Wang
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 |
|---|---|---|---|
| replace SGAN pooling with a graph attention network (GAT) (ablation) | SGAT vs SGAN-20VP-20 · ETH/UCY average, Tpred=8 | ADE: 0.41 -> 0.36 | Improved |
| replace SGAN pooling with a graph attention network (GAT) (ablation) | SGAT vs SGAN-20VP-20 · ETH/UCY average, Tpred=12 | ADE: 0.61 -> 0.47 | Improved |
| replace SGAN pooling with a graph attention network (GAT) (ablation) | SGAT vs SGAN-20VP-20 · ETH/UCY average, Tpred=12 | FDE: 1.21 -> 0.92 | Improved |
| add G-LSTM to model temporal correlations of interactions (ablation) | STGAT-20V-20 over SGAT · ETH/UCY average, Tpred=8 | ADE: 0.36 -> 0.31 | Improved |
| add G-LSTM to model temporal correlations of interactions (ablation) | STGAT-20V-20 over SGAT · ETH/UCY average, Tpred=12 | ADE: 0.47 -> 0.43 | Improved |
| add G-LSTM to model temporal correlations of interactions (ablation) | STGAT-20V-20 over SGAT · ETH/UCY average, Tpred=12 | FDE: 0.92 -> 0.83 | Improved |
| add variety loss (k=20 outputs instead of 1) (ablation) | STGAT · ETH/UCY average (Tpred=8, 12) | ADE: reduced by 22.6% and 16.3% | Improved |
| add variety loss (k=20 outputs instead of 1) (ablation) | STGAT · ETH/UCY average (Tpred=8, 12) | FDE: reduced by 27.4%, 26.5% | Improved |
| add G-LSTM (ablation) | STGAT over SGAT · CUDA memory in training (MB) | memory: 1133 -> 7503 | Hurt |
| add G-LSTM (ablation) | STGAT over SGAT · Inference time (seconds, avg) | time: 11.565 -> 11.680 | No clear effect |
| replace SGAN pooling with GAT (ablation) | SGAT over SGAN-P · Inference time (seconds, avg) | time: 8.579 -> 11.565 | Hurt |
Later papers that compared with this method
1 later accepted paper reports STGAT as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Spatial-Temporal Consistency Network for Low-Latency Trajectory Forecasting (ICCV 2021 · better on 4 of 4 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
- Social-STGCNN: A Social Spatio-Temporal Graph Convolutional Neural Network for Human Trajectory Prediction (CVPR 2020 · similarity 0.91)
- Spatio-Temporal Graph Transformer Networks for Pedestrian Trajectory Prediction (ECCV 2020 · similarity 0.91)
- A Set of Control Points Conditioned Pedestrian Trajectory Prediction (AAAI 2023 · similarity 0.89)
- A Unified Environmental Network for Pedestrian Trajectory Prediction (AAAI 2024 · similarity 0.87)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · similarity 0.87)
- MS-TIP: Imputation Aware Pedestrian Trajectory Prediction (ICML 2024 · similarity 0.87)
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.
