This repository contains the code for our paper published in IJCAI 2021:

• Spline Positional Encoding for Learning 3D Implicit Signed Distance Fields
• Peng-Shuai Wang, Yang Liu, Yu-Qi Yang, and Xin Tong

The code has been tested on Ubuntu 16.04/18.04, please follow the following instructions to install the requirements.

  conda create --name spe python=3.7
  conda activate spe
  conda install  pytorch==1.5.0 torchvision==0.6.0 cudatoolkit=10.2 -c pytorch
  pip install -r requirements.txt

For the task of SDF reconstruction from a point cloud, SDF regression and image regression, please download the data from this link and then unzip it to the folder data.

For the shape space learning, please download the data from the official website of Dfaust, and extract the meshes with the code provided by Dfaust to the target folder, denoted as <dfaust_folder>. Then download the training and testing data list file from this link and unzip the list file to the folder data. After these 2 steps, run the following command to generate the data for training and testing: python scripts/dfaust.py --root_folder <dfaust_folder>

• Run the following command: bash scripts/run_train_sdf.sh

• Run the following command: python scripts/run_regress_img.py

• Run the following command: bash scripts/run_regress_sdf.sh

• Run the following command: bash scripts/run_shape_space.sh. The training process is relatively slow, we provide the trained weights here.

• Run the following command to test the trained shape space: python scripts/run_sdf_space_test.py