## MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene Understanding    

### Version 2.0 - WACV 2025 LLVM-AD Challenge

**Developed by:**  
Tencent, University of Illinois at Urbana-Champaign, Purdue University, University of Virginia  

### License
This dataset is released under the [CC-BY-NC-3.0 License](https://creativecommons.org/licenses/by-nc/3.0/).

---

### Dataset Structure

```
data/
├── images/
│   ├── FR1/
│   │   ├── photo_forward.jpg
│   │   ├── photo_lef_back.jpg
│   │   ├── photo_rig_back.jpg
│   │   ├── point_cloud_bev.jpg
│   ├── FR2/
│   │   ├── photo_forward.jpg
│   │   ├── photo_lef_back.jpg
│   │   ├── photo_rig_back.jpg
│   │   ├── point_cloud_bev.jpg
│   └── ...
├── train_v2.json
├── val_v2.json
└── test_v2.json
```

### Input Data

The dataset includes:

1. **Image Views:**  
   - **Forward View**: A forward-facing photo of the road scene.  
   - **Back Left/Right Views**: Photos capturing the back left and back right perspectives.  

   Examples:  
   ![Forward View](./data/images/FR1/photo_forward.jpg)  
   ![Left Back View](./data/images/FR1/photo_lef_back.jpg)  
   ![Right Back View](./data/images/FR1/photo_rig_back.jpg)

2. **Point Cloud (BEV):**  
   - A Bird's Eye View (BEV) image generated from the 3D point cloud data.  

   Example:  
   ![BEV](./data/images/FR1/point_cloud_bev.jpg)

**Note:**  
Participants can choose which inputs to use for the challenge. HD map annotations are not included in this dataset version. All data adhere to standards for producing HD maps.

---

### Challenge QAs

1. **Scene Type (SCN):** Identify the type of road scene depicted in the images.  
2. **Point Cloud Quality (QLT):** Assess the quality of the point cloud data for the current road area.  
3. **Intersection Detection (INT):** Determine if the main road features a crossroad, intersection, or lane change zone.  
4. **Lane Count (LAN):** Count the number of lanes on the current road. (*May not apply to all cases*)  
5. **Lane Description (DES):** Describe the attributes of the lanes on the current road. (*May not apply to all cases*)  
6. **Scene Captioning (CAP):** Provide a detailed description of the current driving scene. (*May not apply to all cases*)  
7. **Unusual Object Detection (OBJ):** Identify any unusual objects visible in the images. (*May not apply to all cases*)  
8. **Lane Change Prediction (MOVE):** Predict the ego vehicle's lane change behavior.  (*May not apply to all cases*) 
9. **Speed Prediction (SPEED):** Predict the ego vehicle's speed behavior.  (*May not apply to all cases*) 

---

### Citation

If you use this dataset, please cite the following works:

#### Main Dataset Paper:
```
@inproceedings{cao2024maplm,
  title={MAPLM: A Real-World Large-Scale Vision-Language Benchmark for Map and Traffic Scene Understanding},
  author={Cao, Xu and Zhou, Tong and Ma, Yunsheng and Ye, Wenqian and Cui, Can and Tang, Kun and Cao, Zhipeng and Liang, Kaizhao and Wang, Ziran and Rehg, James M and others},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={21819--21830},
  year={2024}
}
```

#### HD Map Annotation System Reference:
```
@inproceedings{tang2023thma,
  title={Thma: Tencent hd map ai system for creating hd map annotations},
  author={Tang, Kun and Cao, Xu and Cao, Zhipeng and Zhou, Tong and Li, Erlong and Liu, Ao and Zou, Shengtao and Liu, Chang and Mei, Shuqi and Sizikova, Elena and others},
  booktitle={Proceedings of the AAAI Conference on Artificial Intelligence},
  volume={37},
  number={13},
  pages={15585--15593},
  year={2023}
}
```