Idea:
My final project is a continuation of the midterm project to work on tattoo image generation using GAN. After exploring this topic/model and having trained a model on face generation, I am interested in utilizing it to fulfill my interest in creating tattoos. So far I have not found any project similar to this on the internet so I am excited to making something new. The idea is to create an interface that allows users to generate tattoo images, ideally incorporating personalized features like user input and style selection.
Inspirations:
I got interested in GAN as I was reading this article on AI Portraits by a research study from Northeastern University. The project allows users to upload images of their faces and the AI will turn them into their celebrity versions. It is fascinating how machine learning and computer vision can recreate such realistic faces.
As I did more research, I found another cool project called this person does not exist. This site creates a generative artificial face image based on real face images dataset it was trained on.
After seeing these two projects, I want to explore and utilize GAN for my final project.
Tools:
I will be using DCGAN (Deep Convolutional Generative Adversarial Network) suggested by Aven. It is said that DCGAN focuses on using Deep Convolutional networks in place of fully-connected networks. It looks for spatial correlations which means a DCGAN is better fitted for image/video data, whereas a GAN can be applied to wider domains, as the model specifics are left open to be addresses by individual model architectures.
Here are the two models I found in GitHub:
https://github.com/robbiebarrat/art-DCGAN
https://github.com/soumith/dcgan.torch
Dataset:
After browsing through the internet, I was not able to find an existing, clean tattoo image dataset for the purpose of training for the final project. Therefore I will have to collect my own.
Currently, I am thinking of continuing with the method used for the midterm project. The approach is to use the python script called “Google Images Download” with chromedriver which scrapes Google Images based on search keywords/key-phrases on Google Images and download them onto my computer. There is a limitation to the scraping tool as I can only download 500 photos at a time. But if I run this enough times, it should give a substantial amount of images. In the meantime, I will keep looking for better methods.
Goal:
My process would be to collect the tattoo image dataset first and train a DCGAN model on it. It should output generative tattoo images. In terms of presentation, I will create a website similar to thispersondoesnotexist.com where users can see random, AI-generated tattoo images.
If I have enough time, I would like to process with my second goal of allowing users to generate tattoos based on their input sketches. The process would be more complicated and I am not exactly sure about the exact approach. I will explore further if time permitted.