Research map / Label-efficient and robust learning / Recognition and long-tail
Where, What, Why: Towards Explainable Driver Attention Prediction
Yuchen Zhou, Jiayu Tang, Xiaoyan Xiao, Yueyao Lin, Linkai Liu, Zipeng Guo, Hao Fei, Xiaobo Xia, Chao Gou
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 the where (attention map prediction) task to joint training (ablation) | LLada · W3DA Traffic Accident (text explanation) | CIDEr-R: 0.307 -> 1.002 | Improved |
| add the where task to joint training (ablation) | LLada · W3DA Traffic Accident (text explanation) | BLEU-4: 0.289 -> 0.376 | Improved |
| add the what task to a map-only baseline (ablation) | LLada · W3DA | KLdiv: 1.500 -> 1.498 | Improved |
| add the why task on top of what (ablation) | LLada · W3DA | KLdiv: 1.498 -> 1.485 | Improved |
| add what and why tasks to attention-only baseline (ablation) | LLada · W3DA | CC: 0.512 -> 0.527 | Improved |
| use end-to-end LLM-driven LLada instead of ERFNet | attention map prediction · W3DA Normal Driving | KLdiv: 1.979 -> 1.219 | Improved |
| use LLada instead of GazeXplain for joint prediction | explainable driver attention · W3DA Normal Driving | KLdiv: 2.578 -> 1.219 | Improved |
| use end-to-end LLada instead of two-stage ERFNet + LLaVA | textual explanation · W3DA Safety-Critical (text explanation) | CIDEr-R: 0.307 -> 1.233 | Improved |
| train LLada on cross-domain W3DA rather than DR(eye)VE-only models | conventional attention prediction · DR(eye)VE test | KLdiv: 1.35 -> 1.04 | Improved |
| train LLada on W3DA instead of BDDA-only FBLNet | conventional attention prediction · BDDA test | CC: 0.64 -> 0.60 | Hurt |
| train on W3DA with LLada vs DADA-only FBLNet | conventional attention prediction · DADA test | CC: 0.50 -> 0.48 | No clear effect |
| use grayscale mask attention maps instead of heatmap overlays as visual prompt (ablation) | MLLM annotation pipeline · W3DA annotation | annotation quality | Improved |
Closest papers by meaning
- Driver Attention as Competitive Allocation: A Scene--Task-Aware Dual-Branch Framework (NeurIPS 2026 · similarity 0.92)
- DriverGaze360: OmniDirectional Driver Attention with Object-Level Guidance (CVPR 2026 · similarity 0.92)
- DriveSpatial: A Benchmark for Spatiotemporal Intelligence in VLMs for Autonomous Driving (NeurIPS 2026 · similarity 0.87)
- DriveCombo: Benchmarking Compositional Traffic Rule Reasoning in Autonomous Driving (CVPR 2026 · similarity 0.86)
- MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception (CVPR 2025 · similarity 0.86)
- SalM²: An Extremely Lightweight Saliency Mamba Model for Real-Time Cognitive Awareness of Driver Attention (AAAI 2025 · similarity 0.86)
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.
