Augmentation of LiDAR Scenes with Adverse Weather Conditions Using Latent Diffusion Models

Requirements

We provide an available conda environment named ldm:

sh init/create_env.sh
conda activate ldm

The Boreas dataset

You can find instructions to download the dataset in site

In Our Method, we use the sequences:

• boreas-2021-05-06-13-19
• boreas-2021-04-08-12-44
• boreas-2020-12-01-13-26
• boreas-2021-01-26-11-22

Training

• Set data_root in configs/autoencoder/boreas/autoencoder.yaml and configs/ldm/boreas/ldm.yaml

• To train LDM, add path/to/your/autoencoder/ckpt to ckpt_path in configs/ldm/boreas/ldm.yaml

# train an autoencoder
python main.py -b configs/autoencoder/boreas/autoencoder.yaml -t --gpus 0,

# train LDM
python main.py -b configs/ldm/boreas/ldm.yaml -t --gpus 0,

• To resume your training from an existing log directory or an existing checkpoint file, use the flag -r:

# using a log directory
python main.py -b path/to/your/config.yaml -t --gpus 0, -r path/to/your/log/

# or, using a checkpoint 
python main.py -b path/to/your/config.yaml -t --gpus 0, -r path/to/your/ckpt

Pre-trained weights

Pre-trained weights are avilable in site

Generate Augmentation

• To generate augmentation with adverse weather conditions (including postprocessing):

python scripts/generate_snow_data -r path/to/your/ldm/ckpt

• To open the generated .bin file:

import numpy as np

path = path/to/your/generated/file

pcd = np.fromfile(os.path.join(path), dtype=np.float64).reshape((-1, 3))

Evaluating Autoencoder

• Statistical metric:
  • Jensen-Shannon Divergence (JSD)
• Distance metric:
  • Chamfer Distance (CD)

pip install open3d

python scripts/autoencoder_metrics.py -r path/to/your/autoencoder/ckpt

Evaluating Our Method

You can evaluate Our Method in 3D Object Detection using our benchmark in site

Acknowledgement

Our codebase builds on site

Citation

@inproceedings{matteazzi2025augmentation,
  title={Augmentation of LiDAR Scenes with Adverse Weather Conditions Using Latent Diffusion Models},
  author={Matteazzi, Andrea and Arnold, Michael and Tutsch, Dietmar},
  booktitle={International Conference on Neural Information Processing},
  pages={198--212},
  year={2025},
  organization={Springer}
}