ME-PCN: Point Completion Conditioned on Mask Emptiness

[paper] [slides]

ME-PCN is a point completion network that leverages emptiness in 3D shape space. It encodes both the occupied point cloud and the neighboring ‘empty points’. It estimates coarse-grained but complete and reasonable surface points in the first stage, followed by a refinement stage to produce fine-grained sur- face details.

Install

Envrionment & prerequisites

• PyTorch 1.10.1
• CUDA 11.3
• Python 3.9
• Open3D
• PyTorch3D
• Visdom

Compile extension modules:

cd emd
python3 setup.py install
cd expansion_penalty
python3 setup.py install
cd MDS
python3 setup.py install

Usage

Download data and trained models

We include demo model and data in demo folder. Those files are uploaded using Git Large File Storage. To get those file inplace, you will need to install Git LFS before clone this repo. This can be done in Debian using APT:

sudo apt install git-lfs
git lfs install

Export visualization results on realdata

Unzip realdata.zip in demo folder. Run export_realdata.py to get the result.