Research map / Probabilistic and scientific modeling / Molecules and proteins
PIE: A Large-Scale Dataset and Models for Pedestrian Intention Estimation and Trajectory Prediction
Amir Rasouli, Iuliia Kotseruba, Toni Kunic, John K. Tsotsos
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 local context (2x box crop) instead of tight bbox crop to image input (ablation) | PIEint · PIE intention estimation | accuracy: 0.69 -> 0.71 | Improved |
| add bounding box coordinates (motion) to bbox image input (ablation) | PIEint · PIE intention estimation | accuracy: 0.69 -> 0.73 | Improved |
| add bounding box coordinates (motion) to local context image input (ablation) | PIEint · PIE intention estimation | accuracy: 0.71 -> 0.79 | Improved |
| use image-only input instead of location for the LSTM encoder-decoder | LSTMed · PIE intention estimation | accuracy: 0.67 -> 0.60 | Hurt |
| replace vanilla LSTM with LSTM encoder-decoder on box coordinates | LSTM baseline · PIE intention estimation | accuracy: 0.63 -> 0.67 | Improved |
| add predicted pedestrian intention as input (ablation) | PIEtraj · PIE trajectory (1.5s) | MSE: 636 -> 611 | Improved |
| add predicted ego-vehicle speed as input (ablation) | PIEtraj · PIE trajectory (1.5s) | MSE: 636 -> 572 | Improved |
| add both predicted intention and speed (ablation) | PIEtraj · PIE trajectory (1.5s) | MSE: 636 -> 559 | Improved |
| add intention to speed-conditioned trajectory model (ablation) | PIEtraj · PIE trajectory (1.5s) | MSE: 572 -> 559 | Improved |
| use PIEtraj instead of Bayesian LSTM | trajectory prediction · PIE trajectory (1.5s) | MSE: 855 -> 636 | Improved |
| use PIEtraj instead of Bayesian LSTM | trajectory prediction · JAAD trajectory (1.5s) | MSE: 1535 -> 1248 | Improved |
| use softsign instead of tanh activation in LSTMs (ablation) | trajectory and speed streams · PIE trajectory | performance: up to 5% | Improved |
Closest papers by meaning
- PSI: A Benchmark for Human Interpretation and Response in Traffic Interactions (NeurIPS 2025 · similarity 0.90)
- SocialMOIF: Multi-Order Intention Fusion for Pedestrian Trajectory Prediction (CVPR 2025 · similarity 0.87)
- Pedestrian Trajectory Prediction with Missing Data: Datasets, Imputation, and Benchmarking (NeurIPS 2024 · similarity 0.86)
- MS-TIP: Imputation Aware Pedestrian Trajectory Prediction (ICML 2024 · similarity 0.86)
- Social-Transmotion: Promptable Human Trajectory Prediction (ICLR 2024 · similarity 0.85)
- BAT: Behavior-Aware Human-Like Trajectory Prediction for Autonomous Driving (AAAI 2024 · similarity 0.85)
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.
