Reproduce “TSCA-Net: Temporal-Spatial Clique Attention for Interpretable Multimodal Pedestrian Trajectory Prediction” 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
Md Mustafizur Rahman, Guangchao Yang, A F M Abdun Noor, Md Imam Ahasan, Md Mahfuzur Rahman, Md Ariful Islam
TSCA-Net predicts multimodal pedestrian trajectories by introducing learnable temporal gating for goal-history interaction, modeling asymmetric social relationships, and dynamically adjusting its motion decoder's expressiveness based on goal uncertainty. The model achieves an average ADE/FDE of 0.13/0.20 meters on ETH/UCY and 6.95/10.43 pixels on SDD, demonstrating up to 13.3% improvement in ADE over previous methods.
Key insights
- Learnable temporal gating in goal-history interaction significantly improves the precision and temporal coherence of predicted goal distributions by allowing selective weighting of historical data.
- Modeling social interactions with asymmetric pairwise relationships more faithfully captures nuanced pedestrian behaviors, such as avoidance or group formation, than traditional symmetric graph approaches.
- Dynamically adjusting the motion decoder's expressiveness based on goal uncertainty allows for better capture of fine-grained behavioral modes in ambiguous regions, balancing model complexity and overfitting.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.