Skip to content

Latest commit

 

History

History
149 lines (94 loc) · 4.62 KB

README.md

File metadata and controls

149 lines (94 loc) · 4.62 KB
Raw

RobIR: Robust Inverse Rendering for High-Illumination Scenes

Project page | Paper | Data

News

  • [10/08/2024] The complete code has been released.
  • [10/03/2024] Project page has been released.
  • [9/26/2024] RobIR (formerly known as SIRe-IR) has been accepted by NeurIPS 2024. We will release the code these days.

Dataset

In our paper, we use:

We organize the datasets as follows:

├── data
│   | nerf 
│     ├── hotdog
│     ├── lego 
│     ├── ...
│   | robir_dataset
│     ├── truck
│     ├── chessboard
│     ├── ...
│   | blendedMVS
│     ├── bear
│     ├── clock
│     ├── ...
│   | dtu
│     ├── scan83
│     ├── scan118
│     ├── ...

Run

Environment

  • Set up the Python environment
git clone https://github.com/ingra14m/RobIR
cd RobIR

conda create -n robust-ir-env python=3.7
conda activate robust-ir-env

pip install torch==1.13.1+cu116 torchvision==0.14.1+cu116 --extra-index-url https://download.pytorch.org/whl/cu116
pip install pyg-lib torch-scatter torch-sparse torch-cluster torch-spline-conv torch-geometric -f https://data.pyg.org/whl/torch-1.13.0+cu116.html
pip install -r requirements.txt

Stage 1: NeuS (Geometry Prior)

cd neus
python exp_runner.py --gin_file config/blender.gin  # for blender dataset
python exp_runner.py --gin_file config/blendedMVS/neus_bear.gin  # for blendedMVS dataset
python exp_runner.py --gin_file config/dtu/neus_dtu83_toy.gin  # for dtu dataset

The mesh and other useful settings are saved in logs.

Stage 2: BRDF Estimation

  • We provide confs_sg/hotdog.conf for general blender scenes and confs_sg/truck.conf for the truck in our robir dataset.
  • We also provide confs_sg/dtu.conf for general real-world scenes.

If you wanna train other scenes, please change the config files, neus_pretrained_path, data_split_dir and exp_name.

Here we take the blender scene hotdog as an example.

2.1 Train Norm

PYTHONPATH=. python training/exp_runner.py --conf confs_sg/hotdog.conf --neus_pretrained_path neus/logs/blender/hotdog-neus --data_split_dir data/nerf/hotdog --expname hotdog --trainstage Norm

2.2 Train Visibility and Indirect Illumination

PYTHONPATH=. python training/exp_runner.py --conf confs_sg/hotdog.conf --neus_pretrained_path neus/logs/blender/hotdog-neus --data_split_dir data/nerf/hotdog --expname hotdog --trainstage Vis

2.3 Train PBR

PYTHONPATH=. python training/exp_runner.py --conf confs_sg/hotdog.conf --neus_pretrained_path neus/logs/blender/hotdog-neus --data_split_dir data/nerf/hotdog --expname hotdog --trainstage PBR

2.4 Train RVE

PYTHONPATH=. python training/exp_runner.py --conf confs_sg/hotdog.conf --neus_pretrained_path neus/logs/blender/hotdog-neus --data_split_dir data/nerf/hotdog --expname hotdog --trainstage CESR

Results

Albedo

image-20231020012659356

Roughness

image-20231020012659356

Envmap

image-20231020012659356

Relighting

image-20231020012659356

De-shadow

See more in the project page.

Acknowledgments

This work was supported by Key R&D Program of Zhejiang (No.2024C01069). We thank Wenxin Sun for her help in pipeline illustration. We also thank Yuan Liu and Wen Zhou for the constructive suggestions.

BibTex

@article{yang2023sireir,
    title={SIRe-IR: Inverse Rendering for BRDF Reconstruction with Shadow and Illumination Removal in High-Illuminance Scenes},
    author={Yang, Ziyi and Chen, Yanzhen and Gao, Xinyu and Yuan, Yazhen and Wu, Yu and Zhou, Xiaowei and Jin, Xiaogang},
    journal={arXiv preprint arXiv:2310.13030},
    year={2023}
}

This work was built on InvRender and NeuS. Please consider citing these two awesome works.