This repository is an unofficial Pytorch implementation of Neural-Pull: Learning Signed Distance Functions from Point Clouds by Learning to Pull Space onto Surfaces.

# clone this repo
git clone https://github.com/bearprin/NeuralPull_pytorch.git

# create a conda environment 
conda env create -f env.yaml

# activate the new conda environment
conda activate neural-pull

# train and evaluate the with default settings
python train.py

• Put your own pointcloud files in 'npy_data' folder, each pointcloud file in a separate .npy file
  • Data will be processed on loading.
• Also put ground-truth mesh in 'mesh' folder for evaluation

To train the model, run this command:

python train.py --name <experiment name>

Each experiment result will be saved in experiment/experiment name and log evaluation to tensorboard.