Lighting transfer across multiple views through local color transforms
MATLAB code for the CVM paper Lighting transfer across multiple views through local color transforms.
Tested on Ubuntu 18.04.1 LTS with MATLAB R2018b. Change directory to /src, run in MATLAB:
>> main
Demo I/Os
For the lighting transfer demo, we provide as inputs:
- an image pair: source (to be edited), target (with desired lighting)
- visible 3D points in the reconstructed point cloud
- estimated camera matrices
Data source: [Laffont et al. 12]. We DO NOT claim copyrights of these inputs. Please DO NOT re-distribute without further agreement.
The outputs include:
- output_ours: a relit image obtained by propagating local color transforms
- output_warped_target: a naive homography warping of the target image
- output_propagated_color: a naive color propagation of the target pixel correspondences
How to use your own photos
We use off-the-shelf software VisualSfM [Wu 2011], which first applies structure from motion [Wu et al. 2011] to estimate the parameters of cameras and then uses patch-based multi-view stereo (pmvs) [Furukawa and Ponce 2010] to generate a 3D point cloud of the scene.
Follow the instructions on the software website to run on your own photo collection. Note that you may want to modify the option file to specify the timages. The software tries to reconstruct 3D points until image projections of these points cover all the target images.
Please copy the pmvs result folder under
/src. The demo_data folder is set to the same folder structure for your reference:
.
├── ...
├── src
│ ├── demo_data
│ ├── models
│ ├── option-custom.patch
| └── ...
│ ├── txt
│ ├── visualize
│ └── ...
│ └── ...
└── ...
Change the following lines of code:
main.m:
line 24| folder_name = 'your_data_folder';
load_3d_points.m
line 4| mode = 'pmvs';
line 13| filepath = fullfile(folder_name, 'models', 'your_custom_option.patch');
Citation
Please cite the paper if you used this code for research:
@article{zhang2017lighting,
title={Lighting transfer across multiple views through local color transforms},
author={Zhang, Qian and Laffont, Pierre-Yves and Sim, Terence},
journal={Computational Visual Media},
volume={3},
number={4},
pages={315--324},
year={2017},
publisher={Springer}
}
Contact
qian_zhang AT brown DOT edu, if you have any questions.