Mesh Total Generalized Variation for Denoising [paper]

:zap:Status Update: [2021/06/08] This paper has been accepted by the IEEE Transactions on Visualization and Computer Graphics (TVCG).

<p align='center'> <img src='images/figure1.png' width='850'/> </p> <!--  -->

by Zheng Liu, Yanlei Li, Weina Wang, Ligang Liu, and Renjie Chen(Corresponding author)

:bulb: Introduction

MeshTGV is an efficient numerical framework to discretize TGV over triangular meshes. Based on this discretization, a vectorial TGV regularization model is proposed to restore the face normal field. Then, we introduce an efficient and effective algorithm to solve the optimization problem.

<p align='center'> <img src='images/figure2.png' width='850'/> </p>

:wrench: Usage

clone this repository, you will get :

MeshTGV
 │
 │── data                  [//comparison data used in our paper]
 │     │—— CAD
 │     │── Kinect
 │     │── NonCAD
 │
 │—— MeshTGV-1.2.0.exe     [//executable file to run our algorithm]
 │—— README.md

run the program :

1. double click the executable file.

2. load a mesh file (.obj) :

   Menu → Model → Load Mesh → Choose your noisy mesh

3. open the MeshTGV algorithm panel :

   Menu → Denoising → MeshTGV

4. set parameters then click the Run button.

5. get output :

   • Visualization output : showed in the mainwindow.
   • Norm result output : automaticlly saved in the same folder with the input Noisy Mesh.
   • Mesh result output : You need to save the Mesh result manually : Menu → Model → Save

parameters setting :

Please see Section 6.1 Parameters Setting in our paper for more details of parameters.

:moyai: About data

There are four files in each model's folder. For example, in Block's folder :

 data                  [//comparison data used in our paper]
   │—— CAD
   │    │—— Block
   │    │    │—— groundtruth.obj
   │    │    │—— noisy.obj
   │    │    │—— tgv_filtered_normals-aad=xxx.txt
   │    │    │—— tgv_result_mesh.obj
   │    │—— ...
   │
   │── Kinect
   │── NonCAD

• groundtruth.obj : The GroundTruth data of model.

• noisy.obj : The Noise data of model.

• tgv_filtered_normals-aad=xxx.txt : The result normals saved after MeshTGV Normal Filtering but before Vertex Updating. "aad=xxx" in the filename means the Average Angle Deviation(θ) of the result normals is "xxx".

• tgv_result_mesh.obj : The result mesh saved after MeshTGV Normal Filtering and Vertex Updating.

:link: Citation

If you find this work helpful please consider citing it :

@article{Liu2021MeshTGV,
  title={Mesh Total Generalized Variation for Denoising},
  author={Liu, Zheng and Li, Yanlei and Wang, Weina and Liu, Ligang and Chen, Renjie},
  journal={IEEE Transactions on Visualization and Computer Graphics},
  year={2021},
  doi={10.1109/TVCG.2021.3088118},
}