README.md

ReSimAD

└── src
    ├── __init__.py
    ├── config # Storing data files for emulator simulation
    │   ├── object_box_db.json # carla blueprint data files (individual vehicle and pedestrian blueprint names, bounding box size, etc.)
    │   ├── sequence_index.txt
    │   └── vec_track # Intermediate Data Files
    │       └── *.pkl
    ├── preprocess
    │   ├── __init__.py
    │   ├── create_tracklets_database.py # Generating Intermediate Data Files for Simulation Using Waymo data
    │   ├── trans_seg_to_sequence.py # Get the name of the mesh imported by the current carla
    │   └── waymo_converter.py # Generate kitti format datasets by converting acquired point cloud data with intermediate data files
    ├── scene_replay.py # WaymoSim simulation data generation pipeline
    ├── sensor_generator.py
    ├── walker_control.py
    └── waymo_sequence # Storing waymo raw data
        └── *.tfrecord

Requirements

• Carla (>=0.9.12)
• PythonAPI for Carla
• requirements.txt

Prepare mesh for CARLA

• Follow StreetSurf to generate Scenario Meshes.
• Thanks to Jianfei Guo for Technical support!!!
• Follow the instruction Import a Large Map into CARLA to import reconstruct mesh file(.fbx)

Build LiDAR

Follow LiDAR Simulation Library to use build different LiDAR.

Generate simulated data

1. Prepare intermediate pkl file

Organize the raw Waymo files(*.tfrecord) in `waymo_sequence` folder. 

Generate the data infos by running the following command:

```bash
python preprocess/create_tracklets_database.py
```

2. Simulated in CARLA

In [lidar_sensor python file](ReSimAD/src/sensor_generator.py), 
- `local_save_dir` *line 53*: change the lidar data save path
- `lidar_name` *line 58*: change different LiDAR, such as: "hdl64", "hdl32", "pandar_40p" ...
- `lidar_transform` *line 62*: change the attachment location of LiDAR.

Generate Lidar data infos by running the following command:
```bash
python scene_replay.py --waymo_sequence segment-9320169289978396279_1040_000_1060_000
```

3. Convert lidar data to Kitti like using the annotations from intermediate file.

```bash
python preprocess/waymo_converter.py
```

The demo fbx and pkl file can be access from google drive

4. Zero-shot domain adaptation

Please refer to 3DTrans for detail zero-shot cross-dataset adaptation results.