# dl-scene

**Repository Path**: ryan.van/dl-scene

## Basic Information

- **Project Name**: dl-scene
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2025-05-17
- **Last Updated**: 2025-05-18

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# SCENE: Reasoning about Traffic Scenes using Heterogeneous Graph Neural Networks

Official repository of the paper:\
**[SCENE: Reasoning about Traffic Scenes using Heterogeneous Graph Neural Networks](https://arxiv.org/abs/2301.03512)**\
Thomas Monninger*, Julian Schmidt*, Jan Rupprecht, David Raba, Julian Jordan, Daniel Frank, Steffen Staab and Klaus Dietmayer\
*Thomas Monninger and Julian Schmidt are co-first authors. The order was determined alphabetically.\
\
IEEE Robotics and Automation Letters (RA-L), 2023

<img src="images/overview.png" width="60%">

The repository contains the source code of our graph convolution operator and our experiments on publicly available knowledge graph datasets.

## Citation
If you use our source code, please cite:
```bibtex
@Article{monningerschmidt2023scene,
  author={Monninger, Thomas and Schmidt, Julian and Rupprecht, Jan and Raba, David and Jordan, Julian and Frank, Daniel and Staab, Steffen and Dietmayer, Klaus},
  journal={IEEE Robotics and Automation Letters}, 
  title={SCENE: Reasoning About Traffic Scenes Using Heterogeneous Graph Neural Networks}, 
  year={2023},
  volume={8},
  number={3},
  pages={1531--1538},
  doi={10.1109/LRA.2023.3234771}}
```

## License
<a rel="license" href="http://creativecommons.org/licenses/by-nc/4.0/">
<img alt="Creative Commons License" style="border-width:0" src="https://i.creativecommons.org/l/by-nc/4.0/88x31.png"
 /></a><br />SCENE is licensed under <a rel="license" href="http://creativecommons.org/licenses/by-nc/4.0/"
 >Creative Commons Attribution-NonCommercial 4.0 International License</a>.
 
Check [LICENSE](LICENSE) for more information.

## Installation
### Install Anaconda
We recommend using Anaconda.
The installation is described on the following page:\
https://docs.anaconda.com/anaconda/install/linux/

### Install Required Packages
```sh
conda env create -f environment.yml
```

### Activate Environment
```sh
conda activate scene
```

## Generate Results
```sh
python3 main.py --dataset=aifb
```
Options for `--dataset` are:
- `aifb`: AIFB知识图谱
- `mutag`: MUTAG分子图  
- `bgs`: BGS地质数据集
- `am`: 亚马逊产品图
- `nuscenes`: NuScenes交通场景数据集

### NuScenes数据集特别说明

#### 1. 安装开发工具包
```bash
# 安装最新稳定版
pip install nuscenes-devkit

# 或安装特定版本(推荐)
pip install nuscenes-devkit==1.1.10
```

#### 2. 数据集下载与准备
**方法一：使用官方脚本下载**
```python
from nuscenes import NuScenes

# 下载迷你数据集(推荐首次使用)
NuScenes.download('v1.0-mini', './data/nuscenes')

# 下载完整数据集(需先申请权限)
NuScenes.download('v1.0-trainval', './data/nuscenes')
```

**方法二：手动下载**
1. 访问[NuScenes官网](https://www.nuscenes.org)注册并申请权限
2. 下载以下文件到`./data/nuscenes`:
   ```
   v1.0-trainval_meta.tgz    # 元数据(必须)
   v1.0-trainval0*_blobs.tgz # 数据文件(共7个)
   ```
3. 解压文件:
```bash
# 单个文件解压
tar -xzvf v1.0-trainval_meta.tgz -C ./data/nuscenes

# 批量解压
for f in v1.0-trainval*.tgz; do tar -xzvf "$f" -C ./data/nuscenes; done
```

#### 3. 数据验证
```python
from nuscenes import NuScenes

# 验证数据集
nusc = NuScenes(version='v1.0-mini', dataroot='./data/nuscenes')
print(f"数据集加载成功! 包含:")
print(f"- {len(nusc.sample)} 个样本")
print(f"- {len(nusc.sample_data)} 个传感器数据")
```

#### 4. 目录结构要求
```
./data/nuscenes/
   ├── v1.0-mini/       # 或 v1.0-trainval
   │   ├── maps/        # 地图数据
   │   ├── samples/     # 关键帧数据  
   │   ├── sweeps/      # 连续帧数据
   │   └── v1.0-mini.json  # 元数据文件
```

#### 5. 注意事项
- 完整数据集需要约300GB空间
- 下载需要注册并通过申请
- 首次运行会构建缓存，约需30分钟
- 推荐使用迷你数据集(v1.0-mini)进行测试

## Results
Results are stored in the `results/` folder.
By default, it contains the original results obtained on our test system.\
Values are reported in [our paper](https://arxiv.org/abs/2301.03512).\
Test system specifications: Intel Core i9-7920X, NVIDIA GeForce RTX 2080 Ti.