Deep Poisson Reconstruction

This is the official implementation of "GradNet: Unsupervised Deep Screened Poisson Reconstruction for Gradient-Domain Rendering" (SIGGRAPH Asia 2019). For more details, please refer to our paper.

Data

All data (dataset, testing scenes and model weights) can be downloaded from here (access code: r71f).
The gradnet_dataset.zip is a multi-part archive. You can merge all parts by zip -F gradnet_dataset.zip --out full_dataset.zip, and then uncompress the full_dataset.zip.

Usage

Train

Make sure the dataset is located in ./dataset.
Create a directory ./saved_models to save model weights.
Run python train.py.

Test

Make sure the testing scenes are located in ./test_data.
Make sure ./saved_models contains the model weight to test.
Run python eval.py --epoch <epoch>.

Citation

If you find it useful in your research, please kindly cite our paper:

@article{GradNet_SA2019,
author = {Guo, Jie and Li, Mengtian and Li, Quewei and Qiang, Yuting and Hu, Bingyang and Guo, Yanwen and Yan, Ling-Qi},
year = {2019},
month = {11},
pages = {1-13},
title = {GradNet: unsupervised deep screened poisson reconstruction for gradient-domain rendering},
volume = {38},
journal = {ACM Transactions on Graphics},
doi = {10.1145/3355089.3356538}
}

Contact

If you have any questions, please feel free to contact guojie@nju.edu.cn.

iRedBean/Deep-Poisson-Reconstruction