This repository contains the official code to reproduce the results from the paper:
PolyNet: Polynomial Neural Network for 3D Shape Recognition with PolyShape Representation (3DV 2021)
[project page] [arXiv] [ResearchGate] [presentation]
- Python 3.8.5
- PyTorch 1.7.1
- numpy
- Pillow
- torch_scatter
Download the preprocessed ModelNet dataset with PTQ and √3-subdivision from the follwing link and unzip them in the data directroy. The data type is .npz
.
[PTQ] [√3-subdivision]
In config.json
you can set dataset type (ModelNet10 or ModelNet40) and the PolyPool type (PTQ, Sqrt3).
To train PolyNet with the desired dataset and PolyPool, simply run,
CUDA_VISIBLE_DEVICES=0 python train.py --config config.json -t "direction to save the model"
If you find our paper, code, or provided data useful, please consider citing:
@INPROCEEDINGS{9665897,
author={Yavartanoo, Mohsen and Hung, Shih-Hsuan and Neshatavar, Reyhaneh and Zhang, Yue and Lee, Kyoung Mu},
booktitle={2021 International Conference on 3D Vision (3DV)},
title={PolyNet: Polynomial Neural Network for 3D Shape Recognition with PolyShape Representation},
year={2021},
pages={1014-1023},
doi={10.1109/3DV53792.2021.00109}}