Few choices in life are more personal, or more permanent, than picking a tattoo. Getting just the one you want may soon get a little easier thanks to AI.
Tattoodo, a Copenhagen-based startup, has come up with a deep learning-powered app that can help people find the right style and image before they submit to the gun.
More than 20 million tattoo lovers, and the plain curious, flock to Tattoodo’s site each month to peruse its massive collection of ink uploaded by people from around the world.
Previously, these uploads had been laboriously hand-categorized and tagged. Then Tattoodo developers Goran Vuksic and Dennis Micky Jensen — tinkering around during some of their company’s regular hacker hours — had the insight to build a neural network for image recognition and apply it to Tattoodo’s collection of body art.
The duo trained their model on the manually tagged collection with a focus on the style or motif of the tattoos. Now, people simply need to type in a textual description of the style they’re interested in and the algorithm will pull up tattoo pictures relevant to the search.
“When you show an interest in an artist, a particular style or motif, the system figures it out and offers you more suggestions of that type of art,” said Vuksic.
Providing What You Want
Vuksic and Jensen created their neural network using the Caffe machine learning framework running on an NVIDIA DIGITS instance on Amazon Web Services.
DIGITS is a deep learning GPU training system that made it easy for the pair to generate the data set and train their model. Accessing it via AWS put the immense processing power of GPUs at their fingertips, without having to invest in a lot of hardware themselves.
The developers are now on their way to curating more than 350,000 tattoo art images for people to draw inspiration from.
And the results are phenomenal. Whether you want a Japanese-style tattoo with a dragon motif or a mandala in full color, the model will recognize your request and suggest tattoos from the database relevant to your search.
It will even show you artists to follow — an important consideration given that works from more than 50,000 artists are part of the Tattoodo collection.
More than 1 million tattoo enthusiasts use Tattoodo’s app. As this number grows, Vuksic and Jensen plan to apply their AI work to improve the upload process, automatically suggesting styles and tags for the art being posted.
With NVIDIA GPUs facilitating the training process, Vuksic and Jensen are also concentrating on improving precision of the results, including further training the model using incoming data from uploads.
“We are the No. 1 destination for tattoo inspiration,” said Vuksic, who says he plans to keep exploring opportunities in AI to keep it that way.
Spoiler Alert: AI Tries to Write the Next Chapter in ‘Game of Thrones’ Saga
If you missed last week’s episode of our AI Podcast, give it a listen. We spoke to an impatient developer who created an AI to write the next chapter in the Game of Thrones saga.
How to Tune in to the AI Podcast
The AI Podcast is available through iTunes, DoggCatcher, Google Play Music, Overcast, PlayerFM, Podbay, Pocket Casts, PodCruncher, PodKicker, Stitcher and Soundcloud. If your favorite isn’t listed here, email us at firstname.lastname@example.org.
Featured image credit: Jayel Aheram, via Flickr.