Research map / Detection and segmentation / Open-set and few-shot detection
Effective Adaptation in Multi-Task Co-Training for Unified Autonomous Driving
Xiwen Liang, Yangxin Wu, Jianhua Han, Hang Xu, Chunjing Xu, Xiaodan Liang
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 an adaptation stage before fine-tuning | MoCo-v2 on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 10.3 -> 61.2 | Improved |
| Add an adaptation stage before fine-tuning | DenseCL on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 20.0 -> 60.7 | Improved |
| Add an adaptation stage before fine-tuning | SimCLR on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 60.3 -> 60.1 | No clear effect |
| Add learned task-specific CLIP prompts (ablation) | LV-Adapter on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 61.5 -> 61.8 | Improved |
| Replace learned prompts alone with vision-to-language adaptation (ablation) | LV-Adapter on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 61.8 -> 61.3 | Hurt |
| Add language-to-vision alignment of text and visual features (ablation) | LV-Adapter on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 61.8 -> 62.2 | Improved |
| Use learned prompts instead of prompt engineering (ablation) | CLIP-based LV-Adapter on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 61.4 -> 62.2 | Improved |
| Use learned prompts instead of prompt ensembling (ablation) | CLIP-based LV-Adapter on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): 61.7 -> 62.2 | Improved |
| Use 1 adaptation epoch and 35 fine-tuning epochs instead of 12 and 24 (ablation) | Pretrain-adapt-finetune · BDD100K Disjoint-normal | mIoU (SS): 59.5 -> 61.5 | Improved |
| Increase adaptation epochs from 6 to 12 while fixing fine-tuning at 35 epochs (ablation) | Pretrain-adapt-finetune · BDD100K Disjoint-normal | mIoU (SS): 61.3 -> 61.9 | Improved |
| Replace the FPN adapter with alternative adapter designs (ablation) | SimCLR adaptation on BDD100K Disjoint-normal · BDD100K Disjoint-normal | mIoU (SS): FPN 60.1; pre-adapter 60.2; post-adapter 60.2; learnable scalar 59.9; learnable vector 59.8 | No clear effect |
| Add LV-Adapter to single-task models | MaskFormer on BDD100K Disjoint-balance · BDD100K Disjoint-balance | mIoU (SS): 57.1 -> 61.3 | Improved |
Closest papers by meaning
- M3Net: Multimodal Multi-task Learning for 3D Detection, Segmentation, and Occupancy Prediction in Autonomous Driving (AAAI 2025 · similarity 0.87)
- VMT-Adapter: Parameter-Efficient Transfer Learning for Multi-Task Dense Scene Understanding (AAAI 2024 · similarity 0.87)
- Prune2Drive: A Plug-and-Play Framework for Accelerating Vision-Language Models in Autonomous Driving (CVPR 2026 · similarity 0.85)
- H-MBA: Hierarchical MamBa Adaptation for Multi-Modal Video Understanding in Autonomous Driving (AAAI 2025 · similarity 0.85)
- MMTL-UniAD: A Unified Framework for Multimodal and Multi-Task Learning in Assistive Driving Perception (CVPR 2025 · similarity 0.85)
- Holistic Autonomous Driving Understanding by Bird's-Eye-View Injected Multi-Modal Large Models (CVPR 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.
