Research map / Video / Action recognition
Drive&Act: A Multi-Modal Dataset for Fine-Grained Driver Behavior Recognition in Autonomous Vehicles
Manuel Martin, Alina Roitberg, Monica Haurilet, Matthias Horne, Simon Reiss, Michael Voit, Rainer Stiefelhagen
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 interior stream to pose model (Two-Stream) (ablation) | pose-based model · Drive&Act fine-grained activities, test | accuracy: 44.36 -> 45.39 | Improved |
| add car-interior network as third stream (ablation) | Two-Stream · Drive&Act fine-grained activities, test | accuracy: 45.39 -> 46.95 | Improved |
| end-to-end I3D instead of Three-Stream pose model | fine-grained activity recognition · Drive&Act test | accuracy: 46.95 -> 63.64 | Improved |
| fuse all sources by softmax score averaging instead of best single view (ablation) | I3D · Drive&Act fine-grained activities, test | accuracy: 64.98 -> 69.03 | Improved |
| combine NIR cameras vs best single NIR camera (ablation) | I3D · Drive&Act fine-grained activities, test | accuracy: 63.64 -> 67.17 | Improved |
| use face view instead of front top view | I3D · Drive&Act fine-grained activities, test | accuracy: 63.64 -> 42.98 | Hurt |
| use I3D instead of pose models at scenario level | scenario/task recognition · Drive&Act test | accuracy: 35.45 -> 31.80 | Hurt |
| train on one view and test on another view | I3D · Drive&Act cross-view | accuracy: 69.57 -> 27.49 (best cross-view) | Hurt |
| I3D vs pose-based models for location classification | atomic action units · Drive&Act test | location accuracy: 56.5 vs 51.12 | Hurt |
Closest papers by meaning
- Toward Driving Scene Understanding: A Dataset for Learning Driver Behavior and Causal Reasoning (CVPR 2018 · similarity 0.89)
- MCAM: Multimodal Causal Analysis Model for Ego-Vehicle-Level Driving Video Understanding (ICCV 2025 · similarity 0.87)
- Panacea: Panoramic and Controllable Video Generation for Autonomous Driving (CVPR 2024 · similarity 0.85)
- MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception (CVPR 2025 · similarity 0.85)
- Driving Video Retrieval for Complex Queries with Structured Grounding (NeurIPS 2026 · similarity 0.85)
- Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes (CVPR 2024 · 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.
