Research map / Video / Video segmentation and tracking
DualScope: Capturing Critical Spatial and Temporal Cues for Distracted Driving Activity Recognition
Zhijie Qiu, Shuaibo Li, Laixin Zhang, Xuming Hu, Wei Ma
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 |
|---|---|---|---|
| replace pretrained ViT-L/14 with behavior-centric encoder BCVE (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 87.23 -> 90.84 | Improved |
| add Dynamic Difference Extractor (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 87.49 -> 90.84 | Improved |
| add Fine-Grained Motion Encoder (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 89.76 -> 90.84 | Improved |
| add SATPM temporal pyramid on top of FGME (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 79.34 -> 90.84 | Improved |
| increase SATPM blocks from 1 to 3 (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 87.53 -> 90.84 | Improved |
| increase SATPM blocks from 3 to 4 (ablation) | DualScope · 5 video datasets avg | top-1 accuracy (%): 90.84 -> 86.42 | Hurt |
| add Gradient Surgery to joint CRCD+IRDD training (ablation) | BCVE · StateFarm+AUC avg | top-1 accuracy (%): 89.31 -> 91.86 | Improved |
| use task-specific adapters instead of full fine-tuning (ablation) | BCVE · StateFarm+AUC avg | top-1 accuracy (%): 90.33 -> 91.86 | Improved |
| use TA+SI adapters instead of LoRA (ablation) | BCVE · StateFarm+AUC avg | top-1 accuracy (%): 90.76 -> 91.86 | Improved |
| add SI-Adapter to TA-Adapter (ablation) | BCVE · StateFarm+AUC avg | top-1 accuracy (%): 90.67 -> 91.86 | Improved |
| apply behavior-centric pretraining (BCVE) in zero-shot | DualScope vs DriveCLIP · SynDD1/StateFarm/DMD/SAM-DD zero-shot | top-1 accuracy (%): 54.00 -> 54.52 (SynDD1); 66.52 -> 66.87 (SAM-DD) | No clear effect |
Closest papers by meaning
- Drive&Act: A Multi-Modal Dataset for Fine-Grained Driver Behavior Recognition in Autonomous Vehicles (ICCV 2019 · similarity 0.86)
- MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception (CVPR 2025 · similarity 0.85)
- Action-slot: Visual Action-centric Representations for Multi-label Atomic Activity Recognition in Traffic Scenes (CVPR 2024 · similarity 0.83)
- DriveGazen: Event-Based Driving Status Recognition Using Conventional Camera (AAAI 2025 · similarity 0.83)
- SalM²: An Extremely Lightweight Saliency Mamba Model for Real-Time Cognitive Awareness of Driver Attention (AAAI 2025 · similarity 0.83)
- Leveraging Driver Field-of-View for Multimodal Ego-Trajectory Prediction (ICLR 2025 · similarity 0.83)
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.
