CUDA-accelerated Signed Distance queries from Meshes

Setup

Tested with CUDA 11.7 (ensure compatible PyTorch version in pkgs-install.sh)

conda create -y -n m2s python=3.8
conda activate m2s
source pkgs-install.sh

Run

Example: Given $DATA_DIR contains meshes as .obj files

cd src/
python main.py \
    --dataset_dir $DATADIR \
    --save_dir $SAVEDIR \
    --num_samples_surf 100000 \
    --num_samples_sdf 600000 \
    --chunk_size 100000

Acknowledgements

Code in this repository is adapted from:

• https://github.com/nv-tlabs/nglod
• https://github.com/zekunhao1995/DualSDF