neusample_release

Code and resources for SIGGRAPH 2023 paper NeuSample: Importance Sampling for Neural Materials

6D Spatially-varying BRDF data

• https://drive.google.com/drive/folders/10vPIMHrnFlLMj7uPatTWRKdARu4_JlhD?usp=sharing
• 40+ different materials for public use including anisotropic and multi-layer; not all are used in the original paper. Credit to Fujun Luan (model training) and Alexandr Kuznetsov, Krishna Mullia (NeuMIP implementation).
• this is in the format of training weights .ckpt, which can be used by Adobe version of simplified NeuMIP implementation (credit to Krishna Mullia and Alexandr Kuznetsov).
• 6D data: 2D for incoming direction; 2D for outgoing direction; 2D for surface uv. Please refer to NeuMIP (original version) and NeuSample (simplified version without mipmap) papers for more details.

Training script examples for various sampling methods

• baselines

  neusample\scripts\train_xs0000_02_baseline.py
  
  neusample\scripts\train_xs0000_05_xie.py

• analytical method:

  neusample\scripts\train_xs0000_00_analytical.py // or neusample\scripts\train_xs0027_00_analytical.py for two material examples.

• normalizing flow:

  neusample\scripts\train_xs0000_04_nsf_prior.py

• histogram mixture:

  neusample\histogram\train_histogram.py

Model inference and render using Mitsuba

  neusample\histogram\eval_mitsuba.py
  
  neusample\histogram\eval_tiled.py (tiled version)

Utilities for visualization

  neusample/scripts/vis_helper.py

Please cite our paper if you find it useful :)

@inproceedings{xu2023neusample,
  title={NeuSample: Importance Sampling for Neural Materials},
  author={Xu, Bing and Wu, Liwen and Hasan, Milos and Luan, Fujun and Georgiev, Iliyan and Xu, Zexiang and Ramamoorthi, Ravi},
  booktitle={ACM SIGGRAPH 2023 Conference Proceedings},
  pages={1--10},
  year={2023}
}

Reference: We built upon https://github.com/VincentStimper/normalizing-flows. Credit goes to them.

Please let us know if you have any questions!

Bing Xu at b4xu@ucsd.edu