Self/Object (2019) explores my personal connection with objects through the creation of experimental portraits in various artistic styles. The objects I chose include a playing card, hourglass, flower, mirror, and spinning top. These self-portraits reflect my feelings and imagination inspired by these objects.

The process of creating these self-portraits in different artistic styles also led me to conduct a style transfer experiment based on my own artworks. While many computational artists have explored using models of famous paintings to create style transfer effects or generate new artworks, I wondered: what if I used the artistic style of my own works for style transfer?

To explore this idea, I selected three of my artworks for the experiment. Below is a video recording of my style transfer website. The website uses a webcam to capture video and applies live style transfer based on my artistic style.

‍

Process:

The self-portraits were photographed and edited by me in Adobe Photoshop. For the web-based fast style transfer, I trained a neural style transfer model on my artworks using a Fast Style Transfer implementation built with deeplearn.js (based on https://github.com/reiinakano/fast-style-transfer-deeplearnjs). After training, I converted the trained model to a browser-compatible format, integrated it into the site, and built the interactive web experience using ml5.js with p5.js, along with HTML and CSS.

‍

Here are my original self-portraits and the image of style transfer results: