/homo_translucent

Primary LanguagePython

Inverse Scattering

This is the implementation for the CVPR 2023 paper:

"Inverse Rendering of Translucent Objects using Physical and Neural Renderers"

By Chenhao Li, Trung Thanh Ngo, Hajime Nagahara

Project page | arXiv | Dataset | Video

Requirements

  • Linux
  • NVIDIA GPU + CUDA CuDNN
  • Python 3
  • torch
  • torchvision
  • dominate
  • visdom
  • pandas
  • scipy
  • pillow

How to use

Test on the real-world objects

python ./inference_real.py --dataroot "./datasets/real" --dataset_mode "real" --name "edit_twoshot" --model "edit_twoshot" --eval

Train

  • Download the dataset.
  • Unzip it to ./datasets/
  • To view training results and loss plots, run python -m visdom.server and click the URL http://localhost:8097.
  • Train the model (change gpu_ids according your device)
python ./train.py --dataroot "./datasets/translucent" --name "edit_twoshot" --model "edit_twoshot" --gpu_ids 0,1,2,3

Test

python ./test.py --dataroot "./datasets/translucent" --name "edit_twoshot" --model "edit_twoshot" --eval

scripts.sh integrate all commands

bash ./scripts.sh

Citation

@inproceedings{li2023inverse,
  title={Inverse Rendering of Translucent Objects using Physical and Neural Renderers},
  author={Li, Chenhao and Ngo, Trung Thanh and Nagahara, Hajime},
  booktitle={Proceedings of the IEEE/CVF Conference on Computer Vision and Pattern Recognition},
  pages={12510--12520},
  year={2023}}

Acknowledgements

Code derived and modified from: