Implementation of the WrappingNet architecture.
The entire framework is illustrated below.
The dataset for WrappingNet should be prepared as follows:
mkdir -p datasets/Manifold40; cd datasets/Manifold40
- Download processed.zip from
https://aspera.pub/3O5IeFo
then move intodatasets/Manifold40/
unzip processed.zip
, then check the data underdatasets/Manifold40/processed/
wget https://cg.cs.tsinghua.edu.cn/dataset/subdivnet/datasets/Manifold40.zip
unzip Manifold40.zip
mv Manifold40 raw
then check the data underdatasets/Manifold40/raw/
pytorch
pytorch-geometric
pytorch-lightning
pytorch-scatter
botorch
open3d
numpy
To use our generalized face convolutions, follow these steps:
- Create a python environment with the above dependencies installed
- Go to
./nndistance/
and runpython build.py install
. This will build the faster chamfer distance module. - Run
CUDA_VISIBLE_DEVICES={GPU}, bash scripts/LC.sh
orCUDA_VISIBLE_DEVICES={GPU}, bash scripts/basesup3.sh
to launch a training script.
Eric Lei, Muhammad Asad Lodhi, Jiahao Pang, Junghyun Ahn, Dong Tian,
"WrappingNet: Mesh Autoencoder via Deep Sphere Deformation",
To Appear in 2024 IEEE International Conference on Image Processing (ICIP).