Culture | Visual Arts

Review: Science Café & Workshop – Can a Machine Replace an Artist? – ‘an exciting glimpse of the partnership between science and art’

11/06/2019

Share on Tumblr0Share on Facebook0Share on Google+0Share on LinkedIn0Tweet about this on TwitterEmail this to someone

Using popular mobile apps like Prisma and DeepArt anyone can produce a piece of stylised art from an input photo, but Daniel Sykora and his research group at the Czech Technical University argue that this wouldn’t pass an artistic version of the Turing test – where even the artist would’ve difficulty judging whether a painting was an original or created by a machine.  So, does this mean sometime in the future artists could be replaced by machines?

In a lively presentation, Skyora argued that artists want to see what they’ve drawn or painted in their own signature style, reproduced authentically using real photos or footage. The current animation techniques used to create movies like Toy Story orFrozen standardise the original images created by the animation artists and don’t reflect the full range and depth of the artist’s creativity. What is the process that could make us believe that an image was the artist’s actual creation?

Sykora reviewed the delivery of neural-based style transfer used in Prisma and DeepArt and found there’s fuzzy colours that don’t resemble the original and the detail is lost. This can’t be used by artists. So what’s essential to preserve the intrinsically unique style of an individual artist? Sykora wants to approach the question from a different perspective with the aim to preserve semantically meaningful details as well as the original richness of artistic media.

Presentation slide 17

Using an artist’s drawing of a ball, the original use of shadow by the artist is transferred to a machine-created image of a golem. Highlights are preserved, as are the shadows. This approach has not been tuned fully into existing style algorithms, so the following image was able to pass the Turing test:

Presentation slide 20

The artist who’d produced the sphere image was asked to draw a Golem figure and six months later was asked if this image was drawn by herself – she wasn’t sure.  Sykora showed a number of other impressive images that also passed the Turing test.

Another advantage of Sykora’s work is that these images can be played with in real time. It’s possible to capture the artist creating an image and applying it to more complex objects. Sykora explained there was a greater challenge once they started to look at more complex images such as faces – to which the human visual system is more sensitively attuned.

Slide 29

Slide 30
The original artist’s image is used to reproduce a photo in the
artist’s original style.  This was then reproduced using statues and Rembrandt portraits with different data – an approach which can be extended to animation. So, instead of the manual process undertaken in the film Loving Vincent movie (Kobiela & Welchman, 2017) for example, this can be done automatically.

Sykora suggests that there would have to be ‘a super challenging program’ for AI to create a completely unique style. The use of ‘style transfer’ is fully dependent on existing original artwork. He admitted that there were Intellectual Property rights issues, which had to be respected. However, most artists, especially those involved in animation, would be happy to work with a tool that increased the scope and productivity of their creative practice. Sykora’s team are currently working with Adobe to innovate film animation. In 2017 Sykora was awarded a Neuron award for young scientists in Computer Science.

The event concluded with audience members using the tools to reproduce their own creativity.  Sykora and his team provided an exciting glimpse of an emerging productive partnership between science and art.

The full slide pack from the presentation can be obtained from the following link: https://dcgi.fel.cvut.cz/home/sykorad/talk_London_2019.pdf

Share on Tumblr0Share on Facebook0Share on Google+0Share on LinkedIn0Tweet about this on TwitterEmail this to someone
facebooktwitter