Based on WNNC(2024 SIGGRAPH Asia), we provide the latest implementations for Gauss surface reconstruction. We also provide the anisotropic surface reconstruction for oriented points (OAGR) which is faster than [PoissonRecon] and has better performance than GR on thin structures.
![Special note: We only provide GPU version code] ![Special note: We only provide GPU version code] ![Special note: We only provide GPU version code]
- For AGR:
cd ext
pip install -e .
cd ..
# width is important for normal quality, we provide a few presets through --width_config
# for clean uniform samples, use l0
python main_AGR.py data/Armadillo_40000.xyz --width_config l0 --tqdm
# for noisy or non-uniform points, use configs l1 (small noise) ~ l5 (large noise) depending on the noise level
# a higher level gives smoother normals and better resilience to noise
python main_AGR.py data/bunny_noised.xyz --width_config l1 --tqdm
...
python main_AGR.py data/bunny_noised.xyz --width_config l5 --tqdm
# the user can also use custom widths:
python main_AGR.py data/bunny_noised.xyz --width_config custom --wsmin 0.03 --wsmax 0.12 --tqdm
# to see a complete list of options:
python main_main_AGR.py -h[Input] If the input is mesh(.ply), the user can change the path ans settings in AGR-ply.py and run:
python AGR-ply.py
[Input] If the input is point(.xyz), the user can change the path ans settings in AGR-xyz.py and run:
python AGR-xyz.py
- For Gauss surface reconstruction:
You may download ANN 1.1.2 and unpack to
ext/gaussrecon_src/ANN.(In fact, compared to WNNC, we have already downloaded and configured usrs. If usrs encounters problems while running WNNC, You can run this code environment, which is compatible with WNNC) Runmakethere. Then go back to the main repository directory, and:
sh build_GR_cpu.sh
sh build_GR_cuda.sh
./main_GaussReconCPU -i <input.xyz> -o <output.ply>
./main_GaussReconCUDA -i <input.xyz> -o <output.ply>For convenience, we provide a Python file for batch reconstruction through AGR. python reconstruction.py