Live Web App: http://paint.test.woza.work/
This is a prototype for an online tool that transforms a portrait photo of a person into a painting. In order to build this app I first had to create a dataset of 10,200 art images using a neural style transfer algorithm. That data was then used to train a U-Net cnn.
The image on the left is the style image that was used in the neural style transfer algorithm.
The process used to build and train the model is described in this Kaggle kernel:
https://www.kaggle.com/vbookshelf/art-by-ai-selfie-painter-web-app
These are the two datasets that were used to train the model:
-
Art by Ai - Neural Style Transfer
https://www.kaggle.com/vbookshelf/art-by-ai-neural-style-transfer -
AISegment.com - Matting Human Datasets
https://www.kaggle.com/laurentmih/aisegmentcom-matting-human-datasets
All javascript, html and css files used to create the web app are available in this repo.