Research map / Probabilistic and scientific modeling / Causal discovery
CaTFormer: Causal Temporal Transformer with Dynamic Contextual Fusion for Driving Intention Prediction
Sirui Wang, Zhou Guan, Bingxi Zhao, Tongjia Gu, Jie Liu
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 RDF, CRE and FSN to a dual-stream Transformer baseline (ablation) | dual-stream Transformer · Brain4Cars [-5,0] | F1-score (%): 95.8 -> 98.6 | Improved |
| add Reciprocal Delayed Fusion (RDF) (ablation) | dual-stream Transformer baseline · Brain4Cars [-5,0] | F1-score (%): 95.8 -> 97.1 | Improved |
| add Counterfactual Residual Encoding (CRE) (ablation) | dual-stream Transformer baseline · Brain4Cars [-5,0] | F1-score (%): 95.8 -> 97.0 | Improved |
| add Feature Synthesis Network (FSN) (ablation) | dual-stream Transformer baseline · Brain4Cars [-5,0] | F1-score (%): 95.8 -> 96.6 | Improved |
| add CRE to RDF+FSN model (ablation) | CaTFormer · Brain4Cars [-5,0] | F1-score (%): 98.0 -> 98.6 | Improved |
| add CRE to RDF+FSN model on the hardest early window [-5,-4] (ablation) | CaTFormer · Brain4Cars truncated [-5,-4] | F1-score (%): 65.6 -> 63.7 | Hurt |
| add RDF to baseline on the earliest window [-5,-4] (ablation) | dual-stream Transformer baseline · Brain4Cars truncated [-5,-4] | F1-score (%): 63.2 -> 61.9 | Hurt |
| use CaTFormer instead of IDIPN with truncated observation windows | CaTFormer · Brain4Cars truncated [-5,-2] | F1-score (%): 74.2 -> 90.1 | Improved |
| use camera-only CaTFormer instead of best prior camera-only baselines | CaTFormer · Brain4Cars (camera only) | F1-score (%): 94.5 -> 97.6 | Improved |
| use Bidirectional Dependency Attention rather than other attention schemes (ablation) | RDF · Brain4Cars | F1-score (%) | Improved |
| set intention-loss weight alpha to 0.10 rather than smaller or larger values (ablation) | CaTFormer loss · Brain4Cars | F1-score (%) | Improved |
Closest papers by meaning
- CausalVAD: De-confounding End-to-End Autonomous Driving via Causal Intervention (CVPR 2026 · similarity 0.88)
- Prioritizing Perception-Guided Self-Supervision: A New Paradigm for Causal Modeling in End-to-End Autonomous Driving (NeurIPS 2025 · similarity 0.86)
- PSI: A Benchmark for Human Interpretation and Response in Traffic Interactions (NeurIPS 2025 · similarity 0.85)
- STaRFormer: Semi-Supervised Task-Informed Representation Learning via Dynamic Attention-Based Regional Masking for Sequential Data (NeurIPS 2025 · similarity 0.84)
- DriveTransformer: Unified Transformer for Scalable End-to-End Autonomous Driving (ICLR 2025 · similarity 0.84)
- DriveGPT: Scaling Autoregressive Behavior Models for Driving (ICML 2025 · similarity 0.84)
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.
