Hans1984 / CineGS

This is the implementation of paper Cinematic Gaussians:  Real-Time HDR Radiance Fields with Depth of Field
MIT License
3 stars 0 forks source link

CinematicGS: Cinematic Gaussians:  Real-Time HDR Radiance Fields with Depth of Field [Pacific Graphics 2024]

Installation

Use the following commands with Anaconda to create and activate your environment:

Note: Our installation method is the same as that of 3D Gaussian Splatting. If you encounter installation issues, please refer to the issues section of it.

Training

For real dataset:

 python train.py -s data_path -m output_path --fd_path .. --ap_path .. --exp_path  .. --length_focal .. --blur

For rendering dataset:

python train.py -s data_path -m output_path --length_focal .. --start_checkpoint pretrained_path/chkpnt7000.pth --blur

please change the base_path to your own dataset path

Rendering

For real dataset:

python render.py -m model_path -s data_path --fd_path .. --ap_path .. --exp_path .. --length_focal ..

For rendering dataset:

python render.py -m model_path -s data_path

The --blur flag should be disabled during rendering.

Post-Editing

python render_taf.py -m model_path -s data_path --fd_path .. --ap_path .. --exp_path  .. --length_focal .. --taf

The aperture size, focus distance, and exposure can be adjusted post-training. You can set appropriate parameters to view the editing results. (The default parameters are for the real dataset scene_3.)

BibTeX

@article{wang2024cinematic,
  title={Cinematic Gaussians: Real-Time HDR Radiance Fields with Depth},
  author={Wang, Chao and Wolski, Krzysztof and Kerbl, Bernhard and Serrano, Ana 
  and Bemana, Mojtaba and Seidel, Hans-Peter and Myszkowski, Karol and Leimk{\"u}hler, Thomas},
  booktitle={Computer Graphics Forum},
  volume={43},
  number={7},
  pages={1--13},
  year={2024},
  organization={Blackwell-Wiley}
}
}

Acknowledge

This source code is derived from the (https://github.com/graphdeco-inria/gaussian-splatting/tree/main), and the depth rendering part is referred to (https://github.com/leo-frank/diff-gaussian-rasterization-depth). We really appreciate the contributions of the authors to that repository.