Awesome

TMNet

This repository contains the source codes for the paper Deep Mesh Reconstruction from Single RGB Images via Topology Modification Networks.

Citing this work

If you find this work useful in your research, please consider citing:

@inproceedings{pan2019deep,
  title={Deep Mesh Reconstruction from Single RGB Images via Topology Modification Networks},
  author={Pan, Junyi and Han, Xiaoguang and Chen, Weikai and Tang, Jiapeng and Jia, Kui},
  booktitle={Proceedings of the IEEE International Conference on Computer Vision},
  pages={9964--9973},
  year={2019}
}

Install

Clone the repo and install dependencies

This implementation uses Pytorch 1.0.0.

## Download the repository
git clone https://github.com/jnypan/TMNet.git
cd TMNet
## Create python env with relevant packages
conda create --name tmnet python=3.7
source activate tmnet
pip install pandas visdom
cd ./extension
python setup.py install

Download

Data

We used the ShapeNet dataset for 3D models, and rendered views from 3D-R2N2:

When using the provided data make sure to respect the shapenet license.

• The ShapeNet point clouds (*.mat) go in data/customShapeNet_mat
• the rendered views (*.png) go in data/ShapeNetRendering

Usage

Train

bash train/train.sh

Generate

python generate.py --model $model_path

Evaluation results

The evaluation is based on 2,048 ground truth points and 2,048 points uniformly sampled from the generated meshes. The CD is multiplied by 1,000 and the EMD is multiplied by 100. The threshold for F-score is set to be 0.002.