How do you create art with AI? Get insights from the Zindi AI art & music challenge winners!

Zindi is excited to announce the winners of the AI Art challenge. Zindians were asked to produce an original piece of visual art or music leveraging AI for a chance to win a trip to AI Expo Africa that took place in Cape Town South Africa 4-5 September. This competition closed on 30 July 2019.

The community stepped up to the challenge with some truly unique and creative artwork. The decision was so difficult that the judges decided to award three winners instead of just two! Congratulations to Matthew Baas (visual art), Emmanuel Onwuegbusi (music) of Nigeria, and Samuel Burnett (video) of South Africa!

The winning art will be auctioned on 3 September at the openning night VIP cocktail at AI Expo Africa. We will see you in Cape Town!

Name: Matthew Baas

Zindi handle: Baas

Where are you from? South Africa

Tell us a bit about yourself.

I grew up in Cape Town, South Africa. I’m an electrical engineering student at Stellenbosch University and I enjoy deep learning, particularly generative models and reinforcement learning.

Tell us about the approach you took.

Most methods for making art with artificial neural networks involve style transfer, where usually a random input image is trained to minimize a style loss and a content loss in order for the image to eventually have the style of one image and the content of another. I wanted to try and swap this usual order around and see if, given as input a stylistic painting (like a Van Gogh painting) as a content image, can we remove the artist’s unique style and leave the fundamental content of the painting behind by using real life photographs as style images? To do this I used a frozen, pretrained VGG19 model for style transfer, using both a pixel MSE loss and perceptual + gram matrix losses from several of the layers in VGG19. What made the greatest difference, however, was a) adjusting the weighting of pixel/perceptual/style loss weightings throughout training; and b) changing the style image regularly to a different photograph, which kept the artwork from latching on to the ‘style’ of a particular photograph’s features.

What were the things that made the difference for you that you think others can learn from?

1 – Google colab is really all one needs to take part in making cool generated images. Without it I would have no had the compute to even attempt this competition.

2 – For style transfer, keep repeatedly display the image at different points during training to see how it is developing, and change your approach accordingly.

3 – The Pytorch website’s tutorials are a really great resource; a lot of the basics of style transfer / neural translation / … are explained quite nicely and simply in their tutorials.

What are the biggest areas of opportunity you see for AI in Africa over the next few years?

The availability of compute for African data scientists. The main hurdle to most data science competitions is the availability and cost of compute, which currently is severely limited in Africa compared to other continents. I think this is a great area of opportunity since things like Google colab and cloud platform providers being more willing and generous to give GPU/compute credits to African data scientists is really democratizing the compute needed to excel in some aspects of the data science field.

What are you looking forward to most about the Zindi community?

The fun of future competitions 🙂

Name: Emmanuel Onwuegbusi

Zindi handle: EmmaMichael

This is Emmanuel’s 2nd Zindi win!

Where are you from? Anambra,Nigeria

Tell us a bit about yourself.

I’m a data scientist and I am passionate about applying artificial intelligence to solving Africa and the world challenges.

Tell us about the approach you took.

My artwork(Music Composition) consists of a Jazz masterpiece created by using Musenet an Openai online tool that uses AI to generate songs with as many as 10 different instruments. citing: Payne, Christine. “Musenet.” OpenAI, 25 Apr. 2019,

What were the things that made the difference for you that you think others can learn from?

The use of a contemporary tool like “Musenet” with availability of choosing different instruments.

Name: Samuel Burnett

Zindi handle: Desktoy

Where are you from? South Africa

Tell us a bit about yourself.

I work as a 3D animator, though I’m an independent game developer in my free time. I would consider myself a 3D generalist.

Tell us about the approach you took.

Most of the effect I used was achieved by compositing different kinds of noises on each other. The colour was created by mapping a gradient to the values of the noise. I created a lot of different versions of this process, some got blurred, some darkened, contrast-ey, etc. Then all those different versions were composited on top of each other with different blending modes. I tried to keep the colour cycles complementary, but mostly just trial and error until the final result looked good. There’s also a separate layer on top of everything that uses a similar process to create subtle chromatic bubbles floating upwards, though in hindsight I could have made that more prominent.

What were the things that made the difference for you that you think others can learn from?

Don’t be afraid to use very vibrant colours – I only went with this look because I accidentally chose a super saturated colour and found it to look better than what I was going for.

What are the biggest areas of opportunity you see for AI in Africa over the next few years, and what are you looking forward to for the Zindi community?

If I were to guess generally, my bet would be on medical diagnoses. I feel as that would have an immense impact on crowded clinics. If it could speed up the process of going through each patient, then more people could be treated.

What are you looking forward to most about the Zindi community?

It’s just great to see these great minds come together – I’m sure I could learn a lot from them.

Please note that use of the image or video presented here is governed by copyright laws of South Africa. They remain the exclusive copyright property of the Creator(s) and Zindi. No rights are granted.

About Zindi

Zindi is the first data science competition platform in Africa. Zindi hosts an entire data science ecosystem of scientists, engineers, academics, companies, NGOs, governments and institutions focused on solving Africa’s most pressing problems.

Zindi works with companies, non-profit organizations, and government institutions to develop, curate, and prepare data-driven challenges. Solutions are ranked automatically by the accuracy achieved. Whether you are testing the data science waters for the first time or trying to crack a persistent business problem with data, Zindi helps organizations push their creative boundaries at an affordable cost.

For data scientists, from newbies to rock stars, Zindi is a place to access African datasets and solve African problems. Data scientists will find all the tools they need on Zindi to compete, share ideas, hone their skills, build their professional profiles, find career opportunities, and have fun!

About AI Expo Africa

AI Expo Africa is the largest business focused Artificial Intelligence (AI) & Data Science B2B trade event and conference in Africa.  Our 2020 conference and expo will run on 3rd-4th September 2020 and builds upon the phenomenal success of the 2018 / 2019 events that were held in Cape Town, that cemented it as the largest gathering of its kind, with over 1000+ registered decision makers, investors, buyers, suppliers, innovators, SMBs and global brands in the region focusing on real world AI and Data Science business applications.