Awesome

Pytorch 1.0 implementation of paper "Patch-base progressive 3D Point Set Upsampling"

This code is a re-implementation of the original tensorflow code in pytorch 1.0. To reproduce results reported in the paper, please use the tensorflow code.

Complete Demo

Make sure cuda, cudnn, nccl is correctly installed. We tested with cuda 9.2 and cudnn 7.4.1 but cuda 9.0+ should work.

# clone
git clone https://github.com/yifita/3PU_pytorch
cd 3PU_pytorch
# download pretrained models
curl -o pytorch_converted_models.zip -O {https://polybox.ethz.ch/index.php/s/QsnhBDg17bX8alE/download}
unzip pytorch_converted_models.zip
# download test data
curl -o data/test_data/test_points.zip -O https://polybox.ethz.ch/index.php/s/wxKg4O05JnyePDK/download
unzip -d data/test_data/ data/test_data/test_points.zip

# conda environment
conda env create -f environment.yml
conda activate pytorch-1.0

# compile
cd losses
python setup.py install
cd ../sampling
python setup.py install

# run code
python main.py --phase test --num_point 312 --num_shape_point 5000 --up_ratio 16 --test_data  "data/test_data/sketchfab_poisson/poisson_5000/*.xyz" --ckpt final_poisson.pth

data preparation

We use h5 file for training, download it here. To create input point clouds from meshes, please follow this instruction.

We converted the pretrained models described here to pytorch. Download them here. This file contains two pytorch binaries final_poisson.pth and final_scan.pth, unzip them

compile

0. Install cuda, cudnn and nccl if you haven't done so yet.
1. Create conda environment conda env create -f enviroment.yml with the prepared enviroment.yml. This will create a conda environment named "pytorch-1.0".
2. Compile pytorch extension

   conda activate pytorch-1.0
   cd losses
   python setup.py install
   cd ../sampling
   python setup.py install
   

execution

testing

# 16x superresolution and save the results under "./model/poisson"
python main.py --phase test --id "poisson" --num_point 312 --num_shape_point 5000 --up_ratio 16 --test_data  "data/test_data/sketchfab_poisson/poisson_5000/*.xyz" --ckpt ./model/final_poisson.pth 

training

Download training data from here.

The training code uses visdom for visualization. Start visdom and go to http://localhost:8097 in browser to see monitor training progress.

conda activate pytorch-1.0
python -m visdom.server

Then run the training command in another terminal

# training using default settings
conda activate pytorch-1.0

python main.py --phase train --id "demo_train" --num_point 312 --num_shape_point 5000 --up_ratio 16 --h5_data train_poisson_310_poisson_625_poisson_1250_poisson_2500_poisson_5000_poisson_10000_poisson_20000_poisson_40000_poisson_80000.hdf5

cite

@InProceedings{Yifan_2019_CVPR,
author = {Yifan, Wang and Wu, Shihao and Huang, Hui and Cohen-Or, Daniel and Sorkine-Hornung, Olga},
title = {Patch-Based Progressive 3D Point Set Upsampling},
booktitle = {The IEEE Conference on Computer Vision and Pattern Recognition (CVPR)},
month = {June},
year = {2019}
}

acknowledgement

This code based is created cortesy of erikwijmans, charlesq34 and yangyan.

contact

Yifan Wang: yifan.wang@inf.ethz.ch