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

Installation

git clone https://github.com/astaka-pe/DeepMeshPrior.git
cd DeepMeshPrior
conda env create -f environment.yml
conda activate dmp

Preparation

mkdir datasets/{c_output,d_output}
mkdir logs/

Mesh Denoising

python denoise.py -i datasets/d_input/dragon

• To view the loss and MAD, run tensorboard --logdir logs/denoise in another terminal and click http://localhost:6006.

Mesh Completion

python completion.py -i datasets/c_input/dragon

• To view the loss, run tensorboard --logdir logs/competion in another terminal and click http://localhost:6006.