This is the official implementation of our TVCG paper "Intrinsic Omnidirectional Image Decomposition with Illumination Pre-Extraction".
To generate the results of our paper quickly:
git clone git@github.com:MrDragon1/OmnidirectionImageIntrinsicDecomposition.git
cd OmnidirectionImageIntrinsicDecomposition
Then run main.m with Matlab (tested on Matlab R2022a). Make sure the current folder contains main.m. The results of Fig.7 and Fig.8 in our paper will be reproduced and saved in img/results folder.
If you want to run with your own image, please the following steps:
- Put the input image into img folder, as well as the depth and segment image.
- Modify main.m and Our.m to make sure the input images are read correctly (.npy or .png are supported here).
- Run main.m, and the results (including reflectance, shading, light source and pre-extraction image) will be saved in img/results folder.
- You can modify the param to run ablation.
Note that you can use other segmentation methods and modify the ceiling_color and floor_color in Our.m to identify the floor and ceiling correctly.