Awesome

Neural Vector Fields (NVF):

Official implementation of Neural Vector Fields (NVF). Feel free to use this code for academic work, but please cite the following:

@misc{yang2023neural,
      title={Neural Vector Fields: Implicit Representation by Explicit Learning}, 
      author={Xianghui Yang and Guosheng Lin and Zhenghao Chen and Luping Zhou},
      year={2023},
      eprint={2303.04341},
      archivePrefix={arXiv},
      primaryClass={cs.CV}
}

Paper - Supplementary - Arxiv - Published in CVPR 2023.

Requirements

The codes have been tested on the linux Ubuntu 20.04 system with NVIDIA RTX3090ti. The enviroment inlcude:

• Python=3.9
• Pytorch=1.12.1
• Torchvision=0.13.1
• CUDA=11.6

Please clone the repository and navigate into it in your terminal, its location is assumed for all subsequent commands.

Installation

The nvf.yml file contains all necessary python dependencies for the project. You can create the anaconda environment using:

conda env create -f nvf.yml
conda activate nvf

There are other dependencies to be installed.

1. To train the model, relavattive operations in point-transfomrer are needed.
2. To extract meshes, meshUDF is needed.
3. To test the model, libaries for chamfer distance, earth mover distance are needed.

The commands have been incorperated by create_env.sh. You can install them via runing the script:

bash create_env.sh

Or you can install step by step by yourself.

Data Preparation

First, create a configuration file in folder configs/, use configs/shapenet_cube_offset_generalization_pt_vq_k16.txt as reference and see configs/config_loader.py for detailed explanation of all configuration options. Change the desired data directory with variable SAVE_DIR in create_split_generalization.py, convert_to_scaled_off.py and boundary_sampling.py. Please also make sure to move the split files split_generalization_*.npz from dataprocessing to your desired SAVE_DIR.

Next, prepare the data for NVF using

python dataprocessing/preprocess.py --config configs/shapenet_cube_offset_generalization_pt_vq_k16.txt

You can generate a random test/training/validation split of the data using

python dataprocessing/create_split.py --config configs/shapenet_cube_offset_generalization_pt_vq_k16.txt

but replacing configs/shapenet_cube_offset_generalization_pt_vq_k16.txt in the commands with the desired configuration.

Training

To train your NVF, you can change the parameters in the configs and run:

python train_generalization.py --config ./configs/${exp_name}.txt 2>&1|tee ${save_dir}/log.txt

In the experiments/ folder you can find an experiment folder containing the model checkpoints, the checkpoint of validation minimum, and a folder containing a tensorboard summary, which can be started at with

tensorboard --logdir experiments/${exp_name}/summary/ --host 0.0.0.0

Generation

To generate meshes after training:

python generation.py

Please specify the desired model before running.

Test

To test results after generation:

python test.py

Please specify the desired experiment name before running. The model for non-watertight shape and garment has been uploaded and you can download it here.

Contact

For questions and comments please leave your questions in the issue or contact Xianghui Yang via email xianghui.yang@sydney.edu.au.

Acknowledge

The code is modified from the NDF. Thanks for open-sourcing.