3D Representation Learning
Advancing Neural Unsigned Distance Fields for Implicit 3D Function Learning
My Contributions
- Enhanced the NDF encoder bottleneck with vector quantization and a learned codebook to capture features in both continuous and discrete modes.
- Trained a modified NDF model on the ShapeNet Cars dataset using the Uni-Siegen OMNI GPU cluster (runtime: 5 days).
- Evaluated generated 3D models with generation metrics, pushing forward implicit 3D representation learning.
Context
This project builds on the seminal work Neural Unsigned Distance Fields for Implicit Function Learning by Julian Chibane, Aymen Mir, and Gerard Pons-Moll, published at NeurIPS 2020.
Original Paper (PDF) | Supplementary | Project Website | Arxiv |
Citation
If referencing this project, please cite the original NDF paper and GIFS paper for vector quantization:
@inproceedings{chibane2020ndf,
title = {Neural Unsigned Distance Fields for Implicit Function Learning},
author = {Chibane, Julian and Mir, Aymen and Pons-Moll, Gerard},
booktitle = {Advances in Neural Information Processing Systems (NeurIPS)},
month = {December},
year = {2020},
}
@inproceedings{ye2022gifs,
title={GIFS: Neural Implicit Function for General Shape Representation},
author={Ye, Jianglong and Chen, Yuntao and Wang, Naiyan and Wang, Xiaolong},
booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
year={2022}
}