This repository is for our Pattern Recognition (PR) 2023 paper 'A Noising-Denoising Framework for Point Cloud Upsampling via Normalizing Flows'. In this paper, we present a novel noising-denoising framework for 3D point cloud upsampling (3DPU), which aims to generate dense points from a sparse input point cloud.

Install the common dependencies from the requirements.txt file

pip install -r requirements.txt

We provide the pre-processed supervised and self-supervised data for the following datasets:

• PU-GAN (Extracting code: 2k2c)
• PU1K (Extracting code: cbhy)
• Sketchfab (Extracting code: 393s)

Please put the datasets in ./data. You can put the datasets elsewhere if you modify the corresponding paths in the args.py.

The directory structure of our project looks like this:

│
├── data                   <- Project data
│   └── PU-GAN 
│   │   └── pointclouds
│   │   │   └── train
│   │   │   └── test
│   │   └── meshes 
│   │   │   └── train
│   │   │   └── test            
│   └── PU1K       
│   │   └── pointclouds
│   │   │   └── train
│   │   │   └── test
│   │   └── meshes 
│   │   │   └── test  
│   └── Sketchfab  
│   │   └── pointclouds
│   │   │   └── train
│   │   │   └── test
│   │   └── meshes 
│   │   │   └── train
│   │   │   └── test      

Current settings in args.py are tested on one NVIDIA GeForce RTX 3090. To reduce memory consumption, you can set batch_size, or patch_size to a smaller number.

Train model on PU-GAN or Sketchfab dataset:

python train.py

Train model on PU1K dataset:

python train_pu1k.py

Test model on PU-GAN, PU1K or Sketchfab dataset:

python test.py

If you find our code or paper useful, please cite

@article{HU2023109569,
  title     = {A Noising-Denoising Framework for Point Cloud Upsampling via Normalizing Flows},
  author    = {Xin Hu, Xin Wei and Jian Sun},
  journal = {Pattern Recognition},
  volume  = {140},
  pages   = {109569},
  issn    = {0031-3203},
  year      = {2023}