NavMapFusion

[Accepted to WACV 2026] NavMapFusion: Diffusion-based Fusion of Navigation Maps for Online Vectorized HD Map Construction

Introduction

This repository is the official implementation of NavMapFusion. The code will be released soon. Stay tuned!

Getting Started

1. Environment

Step 1. Create conda environment and activate it.

conda create --name navmapfusion python=3.8 -y
conda activate navmapfusion

Step 2. Install PyTorch.

pip install torch==1.9.0+cu111 torchvision==0.10.0+cu111 torchaudio==0.9.0 -f https://download.pytorch.org/whl/torch_stable.html

Step 3. Install MMCV series.

# Install mmcv-series
pip install mmcv-full==1.6.0
pip install mmdet==2.28.2
pip install mmsegmentation==0.30.0
git clone https://github.com/open-mmlab/mmdetection3d.git
cd mmdetection3d
git checkout v1.0.0rc6 
pip install -e .

Step 4. Install other requirements.

pip install -r requirements.txt

2. Data Preparation

Step 1. Download NuScenes dataset to ./datasets/nuScenes.

Step 2. Generate annotation files for NuScenes dataset.

python tools/nuscenes_converter.py --data-root ./datasets/nuScenes --newsplit

Step 3. Prepare Navigation Map (OSM) Data.

We follow P-MapNet to prepare the OpenStreetMap data.

Please download the pre-processed data and place it in ./datasets/osm.

3. Training and Validating

To train a model with 8 GPUs:

bash tools/dist_train.sh ${CONFIG} 8

To validate a model with 8 GPUs, an $\eta$ parameter of 0.5, 5 DDIM sampling steps, and a query threshold of 0.5:

bash tools/dist_test.sh ${CONFIG} ${CEHCKPOINT} 8 --eta=0.5 --sampling_timesteps=5 --query_threshold=0.5 --eval

Results

Results on NuScenes newsplit

Model	$\mathrm{AP}_{ped}$	$\mathrm{AP}_{div}$	$\mathrm{AP}_{bound}$	$\mathrm{AP}$	Config	Epoch
StreamMapNet	24.8	18.4	25.6	22.9	Config	24
StreamMapNet-MCA	28.4	22.2	27.8	26.1	Config	24
MapDiffusion	23.4	21.6	22.2	22.4	Config	24
NavMapFusion	32.1	20.7	28.9	27.2	Config	24

Citation

If you find our paper or codebase useful in your research, please give us a star and cite our paper.

@article{monninger2025navmapfusion,
  title   = {NavMapFusion: Diffusion-based Fusion of Navigation Maps for Online Vectorized HD Map Construction},
  author  = {Monninger, Thomas and Zhang, Zihan and Staab, Steffen and Ding, Sihao},
  journal = {arXiv preprint arXiv:2512.03317},
  year    = {2025},
}

Acknowledgments

We sincerely thank the open-sourcing of these works where our code is based on: StreamMapNet, P-MapNet.