Research map / Label-efficient and robust learning / Recognition and long-tail
DiverseAD: A Large-Scale Driving Dataset with Diverse Atmospheric Conditions
Haoyu Wang, Baorui Ma, Donglin Di, Suhang Xuan, Hao Li, Shiliang Zhang
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 driving-intent anchored loss (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.39 -> 1.24 | Improved |
| use direction-based intent similarity instead of ADE (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.38 -> 1.24 | Improved |
| add scene-structure anchored loss (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.42 -> 1.24 | Improved |
| use DINOv3 features instead of Gram-matrix similarity (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.35 -> 1.24 | Improved |
| use soft-weighted KL instead of contrastive loss on structure (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.33 -> 1.24 | Improved |
| use soft-weighted KL instead of contrastive loss on intent (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.38 -> 1.24 | Improved |
| use specialized scene structure expert instead of CLIP (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.37 -> 1.24 | Improved |
| use specialized scene structure expert instead of DINOv2 (ablation) | Ours · DiverseAD | Avg. L2 (m): 1.29 -> 1.24 | Improved |
| fine-tune on DiverseAD | UniAD · DiverseAD | Avg. L2 (m): 31.17 -> 2.11 | Improved |
| fine-tune on DiverseAD | FSDrive · DiverseAD | Avg. L2 (m): 25.50 -> 1.44 | Improved |
| use atmospheric-invariant model instead of fine-tuned FSDrive | Ours · DiverseAD | Avg. L2 (m): 1.44 -> 1.24 | Improved |
| use atmospheric-invariant model instead of OmniDrive | Ours · nuScenes | Avg. L2 (m): 0.40 -> 0.39 | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 2 of 2 on most shared results.
- PPAD — PPAD: Iterative Interactions of Prediction and Planning for End-to-end Autonomous Driving (ECCV 2024) · better on 1 of 1 shared result
- ST-P3 — ST-P3: End-to-End Vision-Based Autonomous Driving via Spatial-Temporal Feature Learning (ECCV 2022) · better on 1 of 1 shared result
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Ithaca365: Dataset and Driving Perception Under Repeated and Challenging Weather Conditions (CVPR 2022 · similarity 0.87)
- Generalized Predictive Model for Autonomous Driving (CVPR 2024 · similarity 0.87)
- BDD100K: A Diverse Driving Dataset for Heterogeneous Multitask Learning (CVPR 2020 · similarity 0.86)
- LargeAD: Large-Scale Cross-Sensor Data Pretraining for Autonomous Driving (TPAMI 2025 · similarity 0.85)
- SearchAD: Large-Scale Rare Image Retrieval Dataset for Autonomous Driving (CVPR 2026 · similarity 0.85)
- Vista: A Generalizable Driving World Model with High Fidelity and Versatile Controllability (NeurIPS 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.
