Skip to content

Code and datasets for TPAMI 2021 "SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces from RGB Images "

License

Notifications You must be signed in to change notification settings

Gorilla-Lab-SCUT/SkeletonNet

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

4 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

SkeletonNet

This repository constains the codes and ShapeNetV1-Surface-Skeleton,ShapNetV1-SkeletalVolume and 2d image datasets ShapeNetRendering. Please download the above datasets at the first, and then put them under the SkeletonNet/sharedata folder.

Prepare Skeleton points/volumes

  • If you want to use our skeletal point cloud extraction code, you can download the skeleton extraction code. This code is built on Visual Studio2013 + Qt.
  • If you want to convert the skeletal point clouds to skeletal volumes, you can run the below scripts.
python sharedata/prepare_skeletalvolume.py --cats 03001627 --vx_res 32
python sharedata/prepare_skeletalvolume2.py --cats 03001627 --vx_res 64
python sharedata/prepare_skeletalvolume2.py --cats 03001627 --vx_res 128
python sharedata/prepare_skeletalvolume2.py --cats 03001627 --vx_res 256

Before running above scripts, you need to change raw_pointcloud_dir and upsample_skeleton_dir used when extracting skeletal points.

Installation

First you need to create an anaconda environment called SkeletonNet using

conda env create -f environment.yaml
conda activate SkeletonNet

Implementation details

For each stage, please refer to the README.md under the Skeleton_Inference/SkeGCNN/SkeDISN folder.

Pre-trained models

We provided pre-trained models of SkeletonNet, SkeGCNN, SkeDISN.

  1. The pre-trained model of SkeletonNet should be put in the folder of ./Skeleton_Inference/checkpoints/all.
  2. The pre-trained model of SkeGCNN should be put in the folder of ./SkeGCNN/checkpoint/skegcnn.
  3. The pre-trained model of SkeDISN should be put in the folder of ./SkeDISN/checkpoint/skedisn_occ.

Demo

  1. use the SkeletonNet to generate base meshes or high-resolution volumes.
cd Skeleton_Inference
bash scripts/all/demo.sh
cd ..
  1. use the SkeGCNN to bridge the explicit mesh recovery via mesh deformations.
cd SkeGCNN
bash scripts/demo.sh
cd ..
  1. use the SkeDISN to regularize the implicit mesh recovery via skeleton local features.
cd SkeDISN
bash scripts/demo.sh
cd ..

Evalation

Please refer to the README.md under the ./SkeDISN folder.

Citation

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

@InProceedings{Tang_2019_CVPR,
author = {Tang, Jiapeng and Han, Xiaoguang and Pan, Junyi and Jia, Kui and Tong, Xin},
title = {A Skeleton-Bridged Deep Learning Approach for Generating Meshes of Complex Topologies From Single RGB Images},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

@article{tang2020skeletonnet,
  title={SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces from RGB Images},
  author={Tang, Jiapeng and Han, Xiaoguang and Tan, Mingkui and Tong, Xin and Jia, Kui},
  journal={arXiv preprint arXiv:2008.05742},
  year={2020}
}

Contact

If you have any questions, please feel free to contact with Tang Jiapeng [email protected] or [email protected]

About

Code and datasets for TPAMI 2021 "SkeletonNet: A Topology-Preserving Solution for Learning Mesh Reconstruction of Object Surfaces from RGB Images "

Topics

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published