Awesome

SAPCU

【Code of CVPR 2022 paper】

Self-Supervised Arbitrary-Scale Point Clouds Upsampling via Implicit Neural Representation

Paper address: https://arxiv.org/abs/2204.08196

Environment

Pytorch 1.9.0

CUDA 10.2

Evaluation

a. Download models

Download the pretrained models from the link and unzip it to ./out/

https://pan.baidu.com/s/1OPVnCHq129DBMWh5BA2Whg 
access code: hgii 

or

https://drive.google.com/file/d/12TifjDW2L7r2LK3AGDgabHdN3f0xY5DR/view?usp=share_link

b. Compilation

Run the following command for compiling dense.cpp which generates dense seed points. Note that the size of the input point cloud is currently limited to 5000, if you want to change this limit, you need to modify the size of the kdtree.

g++ -std=c++11 dense.cpp -O2 -o dense

c. Evaluation

You can now test our code on the provided point clouds in the test folder. To this end, simply run

python generate.py

The 4X upsampling results will be created in the testout folder.

Ground truth are provided by Meta-PU

Training

Download the training dataset from the link and unzip it to the working directory

https://pan.baidu.com/s/1VQ-3RFO02fQfcLBfqvCBZA 
access code: vpfm

or

https://1drv.ms/f/s!AsP2NtMX-kUTml4U3DYUD6Hy9FJn?e=8QfJTH

Then run the following commands for training our network

python trainfn.py
python trainfd.py

Dataset

We present a fast implementation for building the dataset, which is based on occupancy_networks.

a. Preprocessing

Follow the link occupancy_networks to obtain pointclouds and watertight meshes, notice that we only use ShapeNet dataset v1.

b. Building and installing

Then move scripts to occupancy_networks/scripts run the following commands for building traindata for fd and fn:

bash dataset_shapenet/build-fd.sh
bash dataset_shapenet/build-fn.sh
bash dataset_shapenet/installfd.sh
bash dataset_shapenet/installfn.sh

Evaluation

The code for evaluation can be download from:

https://github.com/pleaseconnectwifi/Meta-PU/tree/master/evaluation_code
https://github.com/jialancong/3D_Processing

Citation

If the code is useful for your research, please consider citing:

@inproceedings{sapcu,
  title = {Self-Supervised Arbitrary-Scale Point Clouds Upsampling via Implicit Neural Representation},
  author = {Wenbo Zhao, Xianming Liu, Zhiwei Zhong, Junjun Jian, Wei Gao, Ge Li, Xiangyang Ji},
  booktitle = {Proceedings IEEE Conf. on Computer Vision and Pattern Recognition (CVPR)},
  month     = {June},
  year      = {2022},
  pages     = {1999-2007}
}

Acknowledgement

The code is based on occupancy_networks and DGCNN, If you use any of this code, please make sure to cite these works.