Select Page

GANify: Funhouse Mirror Machine interview broadcast on RTR FM Get Up Mornings

Link: Get Up Mornings [https://rtrfm.com.au/show-episode/getupmorning-2020-01-25/]

Date: 25 January 2020

Interview starts: 01:13:00

Host: Dennis Gedling

Interview participants: Lauren Amos (Creative Producer) and Luke Bennett (Technologist)

Transcript:

— START OF TRANSCRIPT —

DENNIS GEDLING (host)

This is Get Up Morning on RTR FM 92.1

Carnival funhouse mirrors are deeply embedded in our culture, as a source of entertainment, weirdness, self-reflection and in some examples, horror. In the year 2020, however, you might think they’re a little old fashioned. Perth Machine Learning Group and Perth Artifactory have joined forces to prove you wrong. They’ve created GANify: Funhouse Mirror Machine. Funhouse mirrors for the AI era. Have your image transformed by deep learning algorithms, explore the depths of space to see who you really are, or who you could be.

Sound like fun? It’s all happening as part of this year’s Fringe World Festival, and here to tell us more are Lauren and Luke from the Perth Artifactory.

Good morning, guys.

LAUREN AMOS

Morning!

LUKE BENNET

Morning.

DG

Thanks to you for coming into the studio this morning.

First off, I have to ask, why funhouse mirrors?

LA

There are so many ideas, so many exciting ways we could bring machine learning to Fringe and a Perth general audience. Starting out from the developments in computer vision and how we can get computers to recognize things in an image, and transform an image, was really saying: what’s Fringe, what are some of the classic ideas, which you mentioned about funhouse mirrors? You come in and see yourself stretched or warped in a funny sort of way; we can now create some interesting transformations using technology. Updating it for the latest tech.

DG

You mentioned a term there, machine learning, which piqued my interest straight away. Can you explain what you mean by that?

LA

Sure. I think when people think of Artificial Intelligence it’s quite a broad field. And machine learning is one way we try and teach computers to do things, things that humans have been able to do. Machine learning is behind all the exhibits that we’ve done, with slightly different models. It’s basically, how can we tell a computer to recognize an object, or how do we tell a computer to create something from an example set that it’s seen before.

DG

OK. So how does this project work, what are people going to experience when they walk in?

LB

We’ve got three different models. The first one is a GAN model, which basically you’re trying to understand the distribution from a series of previous images, and it translates into images. Basically, you use a webcam to interface with that image, so it creates a new image based on yourself. A second model understands where your body’s positioned relative to itself. So when you’re standing up, it translates that into music, so different tones, a filter on top of that music, based on the position of your body.

DG

So it takes an image and makes sound out of it?

LB

Sort of. It’s two separate models that feed into itself. So in one you stand in front of the model and the model detects where your hands are relative to the rest of your body, and creates (x, y) coordinates within that. Then that translates into a set of filters, and that filter is applied on top of the music itself.

The third show is basically called Career Guidance, so it’s basically a multi-label classification model that segments pieces of an image, a webcam image, and displays it to the user. And that image will have a series of careers that we have defined based on a costume. This costume is worn by the user and the user will get back a career based off that.

DG

[laughs] Fantastic. That sounds like fun.

Are these in like three separate spaces in a room, or how does it work?

LB

Exactly, yeah.

LA

So, we’re at the Girls School, in East Perth. I don’t know if people have checked that out, but we’ve scored one of the old classrooms, which we’ve divided into an interactive space.

DG

[laughing] That seems highly appropriate.

Can you tell us a little bit about, and full disclosure here, I’m not a science or tech head at all, but if you can explain, to my very dumb brain, how this sort of works with the technology you’re using?

LA

I’ll give one explanation. I really love that GANify is based on a GAN, which is a Generative Adversarial Network. Basically, it means we’re trying to teach a computer to come up with a new example of something. Like say we want to get it to draw a cat, and we say, right here’s a bunch of cat pictures, learn a style of cat and come up with a cat. At the same time we teach another bit of the computer to be really good at judging if something is a real cat or a fake cat. So, on one hand you’ve got a computer recognizing bits of… what’s in a data set of cats, and saying I’m going to try and generate this. And then you’ve got the other critic saying, I’m also learning what makes up a cat, and that’s good or that’s not good. And by saying how good it is and what the errors are, it learns to get better over time. So there’s this evolution where there’s the creator and the critic, both learning and producing better quality images.

DG

What sort of applications does this learning and this technology have outside of the Fringe experience?

LA

We’ve used different kinds of machine learning models, some of them have very direct applications. So one of the examples we talked about before was image detection, so that’s recognizing an object within a frame. And that’s used for pothole detection, person recognition, and limitless applications. Some of the generative models I just talked about, they’re great for producing examples of things you haven’t seen before. But maybe at the moment are more creative applications, and we’re looking at how you could translate them into something that’s more business oriented.

LB

Even the actual language processing, so, like, auditory type stuff. There are a few applications with that. Say you want to generate some text or some auditory thing based on something, then it’s very useful for that kind of thing, application.

DG

These technologies sound like they’re already being used quite widely. Is this show maybe an opportunity for people to actually think about how this technology’s being used and things that they maybe use every day?

LA

Definitely. Look, it’s not written in lots of detail how it works in the background, it’s designed to be an engaging person-first, rather than let’s knock you over the head with lots of maths and code. But if people do want to find out a bit more about these technologies, then they can engage with the groups that are behind this. The Perth Machine Learning Group has over 2000 members of its Meetup group, and has multiple events a week. And I guess people need to have fun, but they also have the option to think, wait how does this work, why does this work, and what else might this be used for? So the example of the image detector, we’ve got cameras going up all around Perth that are used to recognize objects. We’re doing it in a fun context here, but what else are they recognizing? Computers are able to do that now, so it’s not just having a person there watching things at the right time. How do people feel about this being used by different organisations?

DG

So it’s a good initiation of that conversation, that thought process, that people might not think about. People don’t, or I don’t, really think about things like that, so this is a really good way to instigate that.

Luke, this is a collaboration between Perth Machine Learning Group and Perth Artifactory, and you’ve worked together before. What does each group bring to this collaboration?

LB

OK, so, Perth Machine Learning Group is much more about the coding side of things. So they’ve helped with the model development. A lot of people from that group have helped with that sort of thing. The Artifactory is much more the physical side of things, so like craft and building up the displays and bringing all that together. The two together have helped with the technology and the crafts, to bring about an art event that interacts well with itself.

DG

And the displays kind of look like old funhouse mirrors, is that the feeling we’re getting?

LA

It’s a bit more of a traditional setting at the moment. We’ve tried to carry on the idea of the Girls School, so we’ve wrapped it in a theme so that the funhouse mirrors is actually a yearbook photo, if you think portrait gallery. We’ve got a little school disco section. But the Artifactory were a great help. They demolished a warehouse to get recycled materials to build out the space, but you might not notice when you go there because it looks quite professional. But really great, both sides coming together, if you like, the hardware and the software. The build and the creative side.

DG

Look, it all sounds incredibly interesting and a lot of fun as well.

Where is it happening, is it happening at the Girls School, can you tell people exactly where that is?

LA

The Girls School, if you haven’t been, it’s the old police headquarters, it was an old girls school, at the end of Wellington Street, number 2. If you want to find us we’re pretty close to the front entrance, just head up the stairs turn right and we’re the door on the right there.

DG

And it’s on every day of the Festival, is it?

LA

Every day. Starts around five-thirty till late.

DG

Fantastic. Look, it sounds like a hell of a lot of fun.

Lauren and Luke, thank you so much for coming in to RTR FM this morning.

That was Lauren and Luke from The Perth Artifactory there, telling us about GANify: Funhouse Mirror Machine, which is running as part of this year’s Fringe World Festival. Head to the Fringe World website to find out more.

— END OF TRANSCRIPT —