Instant recovery of shape from spectrum via latentspace connections

This repository is the official implementation of Instant recovery of shape from spectrum via latentspace connections and Spectral Shape Recovery and Analysis Via Data-driven Connections.

It is now available a PyTorch implementation. Thank you Emilian!

Requirements

To install requirements:

pip install -r ./code/requirements.txt

Training

To train the model with FC encoder:

python ./code/train_dense.py

To train the model with PointNet encoder:

python ./code/train_pointnet.py

The code is tested on: Python 3.6 Tensorflow 2.0

Pretrained models

To download the pretrained models:

python ./models/download_pretrained.py

To download the datas:

python ./data/download_data.py

Evaluation

To replicate the shape from spectrum:

python eval_dense.py

The expected result on pre-trained model is:

	Our	NN
full	1.61e-05	4.47e-05
1000	1.61e-05	4.63e-05
500	1.71e-05	4.01e-05
200	2.13e-05	2.65e-05

To replicate the spectrum from pointcloud on a FLAME shape:

python eval_pointnet.py

The expected result on pre-trained model is:

Citation

If you use our work, please cite our papers.

@article{marin2021spectral,
title={Spectral Shape Recovery and Analysis Via Data-driven Connections},
author={Marin, Riccardo and Rampini, Arianna and Castellani, Umberto and Rodol{\`a}, Emanuele and Ovsjanikov, Maks and Melzi, Simone},
journal={International Journal of Computer Vision},
pages={1--16},
year={2021},
publisher={Springer}
}

@inproceedings{marin2020instant,
title={Instant recovery of shape from spectrum via latent space connections},
author={Marin, Riccardo and Rampini, Arianna and Castellani, Umberto and Rodola, Emanuele and Ovsjanikov, Maks and Melzi, Simone},
booktitle={2020 International Conference on 3D Vision (3DV)},
pages={120--129},
year={2020},
organization={IEEE}
}
  

This work is licensed under a Creative Commons Attribution-NonCommercial 4.0 International License. For any commercial uses or derivatives, please contact us.