Deep Mesh Prior

Learning to Generate 3D Shapes and Scenes | CVPR 2021 Workshop [Paper]

Deep Mesh Prior is an unsupervised mesh restoration method using graph convolutional networks, which takes a single incomplete mesh as input data and directly outputs the reconstructed mesh without being trained using large-scale datasets.

Getting Started

Environments

python==3.8
torch==1.13.0
torch_geometric==2.2.0

Installation

git clone https://github.com/astaka-pe/DeepMeshPrior.git
cd DeepMeshPrior
docker build -t astaka-pe/dmp .
docker run -itd --gpus all --name dmp -v.:/work astaka-pe/dmp
docker exec -it dmp /bin/bash

Preparation

mkdir datasets/c_output datasets/d_output logs/

Mesh Denoising

python3 denoise.py -i datasets/d_input/dragon

Mesh Completion

python3 completion.py -i datasets/c_input/dragon

Follow-up Works

Please check out our newer works, "Dual-DMP" for mesh denoising and "SeMIGCN" for mesh inpainting.