XCI-Sketch: Extraction of Color Information from Images for Generation of Colored Outlines and Sketches
V Manushree, Sameer Saxena, Parna Chowdhury, Manisimha Varma, Harsh Rathod, Ankita Ghosh, Sahil Khose
Official implementation of https://arxiv.org/abs/2108.11554 paper. This paper was accepted at CtrlGen, Machine for Creativity and Design, Deep Generative Models and their Downstream Applications, and New in ML workshops, NeurIPS 2021 .
Checkout our model predictions!
- It is available here on Streamlit Sharing.
Methodology
Our work aims to convert photographic images into colored sketches.
We use the Contour Drawing Dataset presented in the Photo-Sketching paper and propose two ways to extract color information from the images and amalgamate it with the corresponding sketches:
- Rendering Colored Outlines in Sketches:
- We formulate a process to transfer color onto the existing black and white sketches in the dataset to produce colored outline sketches.
- We first select an optimal threshold value (TryingDifferentThresholdValues.ipynb) based on which we perform preprocessing and K-Means Color Clustering (GeneratingImagesBasedOnKVals.ipynb) on the images.
- Using the post-processing image and grayscale sketches we generate colored outlines (GeneratingOutlines.ipynb).
- Generating Colored Sketches:
- We propose another method to produce color-filled sketches by performing colorspace manipulation (data_preprocessing.ipynb).
- We go a step further to use these sketches as the training dataset for a Generative Adversarial Network and develop a model which can generate colored sketches from unseen images (Sketch_gan.ipynb).
Download the data here
License: This work has been apadted from dataset which is licensed under CC BY-NC-SA (Attribution-NonCommercial-ShareAlike). That means you can use this dataset for non-commerical purposes and your adapted work should be shared under similar conditions.