TomoSAR2Height

This repository provides the code for training and evaluating TomoSAR2Height, a method for reconstructing building heights (nDSMs) from spaceborne TomoSAR point clouds.

🛠️ Installation

• Clone the repository

  git clone git@github.com:zhu-xlab/tomosar2height.git
  cd tomosar2height

• All-in-one installation to create a conda environment with all dependencies

  conda env create -f environment.yml && conda activate tomosar2height

• If you prefer manual installation, follow these steps

  conda create --name tomosar2height python=3.10
  conda activate tomosar2height
  conda install pytorch==2.3.0 torchvision==0.18.0 pytorch-cuda=11.8 pytorch-scatter affine laspy matplotlib rasterio scikit-learn scipy tabulate tqdm transformations trimesh urllib3 wandb hydra-core hydra-colorlog omegaconf gdal=3.6 -c pyg -c pytorch -c nvidia -c conda-forge
  pip install open3d==0.18.0

🚀 Usage

📂 Data preparation

Due to licensing restrictions, we cannot redistribute the datasets used in this work. However, you can adapt the repository to your own data by following the steps below.

• Place your own point cloud, building mask, and nDSM raster. A typical directory structure is:

  data/your_dataset/
    cloud/              
      points.las        # point cloud
    raster/
      footprint.tif     # building mask
      ndsm.tif          # nDSM
  

• Create a config file in conf/dataset, using the provided examples as templates. Make sure that the data are defined in the same CRS.

• Preprocess the dataset:

  # Build the Berlin data
  python scripts/build_dataset.py dataset=berlin
  
  # Build the Munich data
  python scripts/build_dataset.py dataset=munich

🎯 Training

Train TomoSAR2Height using different data modalities:

• Using point clouds only

  # Replace `berlin` to `munich` for Munich data
  python train.py dataset=berlin use_cloud=true use_image=false wandb=true run_suffix=_cloud gpu_id=0

• Using point clouds and images

  # Replace `berlin` to `munich` for Munich data
  python train.py dataset=berlin use_cloud=true use_image=true wandb=true run_suffix=_cloud+image gpu_id=0

📊 Evaluation

Before evaluation, make sure checkpoints are available at ./outputs/TomoSAR2Height-{dataset}{run_suffix}/check_points/model_best.pt.

• Evaluate a trained TomoSAR2Height model (point clouds only):

  # Berlin data
  python test.py dataset=berlin use_cloud=true use_image=false run_suffix=_cloud gpu_id=0
  
  # Munich data
  python test.py dataset=munich use_cloud=true use_image=false run_suffix=_cloud gpu_id=0

• Evaluate a trained TomoSAR2Height model (point clouds and images):

  # Berlin data (point cloud & image)
  python test.py dataset=berlin use_cloud=true use_image=true run_suffix=_cloud+image gpu_id=0
  
  # Munich data (point cloud & image)
  python test.py dataset=munich use_cloud=true use_image=true run_suffix=_cloud+image gpu_id=0

Specify run_suffix={YOUR_SUFFIX} with your desired suffix if needed. The results will be saved at ./outputs/TomoSAR2Height-{dataset}{run_suffix}/tiff_test.

⚙️ Available configurations

# check available configurations for training
python train.py --cfg job

# check available configurations for evaluation
python test.py --cfg job

Alternatively, review the configuration file: conf/config.yaml.

⚙️ Pretrained weights

Pretrained weights are available at this link.

🎓 Citation

If you use TomoSAR2Height in a scientific work, please cite the paper:

@article{chen2026tomosar2height,
  author={Chen, Zhaiyu and Wang, Yuanyuan and Shi, Yilei and Zhu, Xiao Xiang},
  journal={IEEE Transactions on Geoscience and Remote Sensing}, 
  title={Reconstructing Building Height from Spaceborne TomoSAR Point Clouds Using a Dual-Topology Network}, 
  year={2026},
  volume={64},
  number={},
  pages={1-15},
  doi={10.1109/TGRS.2026.3656340}
}