Skip to content

Latest commit

 

History

History
71 lines (41 loc) · 2.75 KB

README.md

File metadata and controls

71 lines (41 loc) · 2.75 KB

LoCoPalettes

LoCoPalettes: Local Control for Palette-based Image Editing (project page)

Computer Graphics Forum (CGF). Presented at EGSR 2023.

[By Cheng-Kang Ted Chao, Jason Klein, Jianchao Tan, Jose Echevarria, Yotam Gingold]

See demo video for our editing framework.

About

This repo is official code release for LoCoPalettes.

The contribution of this work:

  1. An approach to compute sparser weights, comprimising spatial coherence with sparsity, to achieve sparse color edits without much color leakage compared to state-of-the-art.
  2. An optimization framework that operates on our proposed palette hierarchy, enabling semantic color editing via placing color constraints.

Installation

You can install dependencies using conda.

Conda

Install Anaconda or Miniconda. (Miniconda is faster to install.) Choose the 64-bit Python 3.x version. Launch the Anaconda shell from the Start menu and navigate to this directory. Then:

conda env create -f environment.yml
conda activate locopalettes

To update an already created environment if the environment.yml file changes, first activate and then run conda env update --file environment.yml --prune.

Usage

First, we need to have features (.mat) files. To extract per-pixel features, please refer to Aksoy's code and run

sh run_extract_feat.sh

Then, we need to extract panoptic segments (followed by guided filtering), run

python panoptic.py <your_image>

This would create a folder named <sss_your_image> that contains segments under root -> classes -> instances hierarchy.

Then, run the below to extract all informations (i.e. palettes, weights, masks, activations, trees):

python func/seg.py <your_image> <your_feature> <sss_your_image> --o s --m fea

The above code generates all informations needed alongside GUI.py,

Finally, run

python GUI.py

After the GUI is loaded, follow the below order clicking:

(1) Load image (2) Load features (3) Load tree (choose arbitrary image from your <sss_your_image> folder)

License

CC BY 4.0

This work is licensed under a Creative Commons Attribution 4.0 International License.