#ICYMI_011: Looking for Truths and Lies in Art and AI (and for Waldo)

What have I done…

AI-Tools like Artbreeder inspire to more experiments. Since people experiments with the re-humanization of statues, I tried my luck with Voltaire. This smirking Monsieur is standing as a bust in the DADA-cradle Cabaret Voltaire, Zurich.

So I tried — and it worked:

And evermore: using First Order Motion Model and a Telegram-Bot “Round DFBot” I could revive him, saying his famous:

We must cultivate our garden (from: "Candide")

Pardon my French pronunciation:

And here as a statue:

Breath ZeroX

If you use AI for multimedia, you need audiovisual (at least). Artbreeder gave me already some beauty. Using AI-generated music by OpenAI (JukeBox) you can produce something never being seen before.

Here is my new series: Breath ZeroX.

Everything is here AI-generated.

But enough about me — there are so many fascinating things in the world.

Artificial Intelligence

01. Day and Night.

A researcher group of Hacettepe University and University of Tübingen developed a new approach for AI-based generation of natural scenes “via Hallucination”: Manipulating Attributes of Natural Scenes via Hallucination (Page / Paper / Code)

Source: Manipulating Attributes of Natural Scenes via Hallucination

Using GAN and style transfer involves a semantical analysis of various landscape attributes. So, changing specific layers, you can change these features, like season or daytime.

On their Project page, you can test some of the functions and also see additional creative abilities of this code.

Let’s keep it in focus.

02. Nightingale or lark?

We know Speech2Face, which enabled “predicting” human faces after just one data: their voice:

Now here is another interesting approach: using birdsong spectrograms to generate images of birds.

Next time Romeo and Julia have an ornithological discussion, they should use this GitHub repository.

03. Learn as much as you GAN

This interactive Colab will explain to you a lot about various Generative adversarial networks.

And we know: Colab Notebooks are awesome.

04. Endless Source of Information

When you are interested in Machine Learning, and especially in Neural Language processing, you should follow Elvis and his NLP Newsletter. So many topics, experiments, datasets, research summaries, and mentions:

And my advice: read the NLP summaries for 2018 and 2019. You will be best informed:

05. Human Errors in Deep Learning.

Bias, unreliability, volatility — people are still uncertain about the abilities of Artificial Intelligence. What if it goes nuts and deliver chaos?

Well, most mistakes in AI have human origin: insufficient labeling of datasets, no consideration of relevant aspects in programming, etc.

Elvis found an excellent paper by Humbatova et al. with a chart

Here is it in big:

Source: Taxonomy of Real Faults in Deep Learning Systems (Nargiz Humbatova, Gunel Jahangirova, Gabriele Bavota, Vincenzo Riccio, Andrea Stocco, Paolo Tonella)

As we see, Deep Learning is a powerful instance — when humans are preparing everything properly.

06. Wake up. Let’s draw electric sheep.

I mean, why not?

Sketch-RNN, a recurrent neural network, is doodling with you. It’s an older one, but amusing.

Here is a sheep I doodled together with RNN:

07. Pixelfie

Do you like Pixel-Art? This tool converts a photo into pixel art. Not more, not less. Provided by Sato (we know him already).

Art

08. AI Art progressing

I already wrote about that great long-term experiment by Peter Baylies. He is training a conditional WikiArt model — now with the help of RunwayML, already since last December.

The results are stunning:

Read more about training Art:

09. This*PLEASENOT*does-not exist.

After StyleGAN emerged, a whole series of non-existing realities were brought to life. Every time you refresh the page, something new appears.

• ThisPersonDoesNotExist.com — StyleGAN2-generated portraits (the most prominent one)
• ThisArtworkDoesNotExist.com — Artwork trained model (by Michael Friesen, I also used this dataset)
• ThisCatDoesNotExist.com — some random cats from the Latent Space
• ThisStartupDoesNotExist.com — a ready page of a startup in an alternate reality — with contact, product description, and even pricing.
• ThisWaifuDoesNotExist.net — a project by the legendary Gwern Branwen, with StyleGAN2-generated Anime characters and GPT-2-generated background story
• ThisChemicalDoesNotExist.com — irreal chemistry

And now, this collection is complemented with… feet.

I am not into feet, but here your mind will be disturbed by animated imagery of feet and a background story about art and science behind AI-generated visuals. Worth reading.

Art

10. Wahrheit und Dichtung

We are living in Fake-News-Epoch. Fact-checkers are the force to liberate us from delusions and fraud in Social.

The problems begin when it’s about art. A photo-collage published by Saatchi-gallery was blocked on Instagram:

Some hair-splitting fact-checker found out, this photo was “not real”, equal “fake”. Well, indeed, it was a photo-manipulation. But the goal of manipulation, in this case, was a creative one. Sure, with all the political, sociological, critical intentions. Or probably without. Why not?

And the question is: do we have to fact-check artworks? Do they tell us the truth? And if yes, how? Should we from now avoid every creative exaggeration or metaphorization of our reality?

The fact-checking is vital as prevention of politically implied lies and deceptions. But art?

What shall we do? Probably in this way?

Of course, this will cancel all the century-long work of Dadaists and Surrealists questioning art. But such a tag could prevent fact-checker workflow?

What do you think?

11. Don’t do it at home, kids.

Another case of content control is circulating especially in the Russian net right now. In May 1945, Red Army photographer Yevgeny Khaldei immortalized the soldier with Red Flag on the roof of Reichstag — an iconic photo symbolizing the end of WWII.

The colored version of this photography was published by various people last Saturday, on the 9th of May 2020 — a holiday that commemorates the surrender of Nazi Germany in 1945.

Well, the posts are mostly banned by Facebook (some users even posted to find out whether they’ll get banned — and their expectations were satisfied).

Why was the photo banned? It depicts dangerous activities: people on the roof. Don’t do it at home, kids.

12. Waiting for Waldo.

You undoubtedly know and love (or hate) this huge panorama of a socio-cultural mess: your task is to look for Waldo, who is hiding like a KGB spy somewhere between two market stands.

An old story: Wally or Waldo, a funny character by the English illustrator Martin Handford was desperately searched and found by generations of kids and adults. Even without knowing, what did he do for not leaving him alone in his overcrowded Hermitage?

We have two new approaches now.

1st one: use Data Science!

They did it already centuries ago, 2017, compare this research, published in HackerNoon:

Jonathan Fly also uses U2-Net salience detection model with convincing results:

But in times of Social Distancing (which is ridiculous), we have to rethink the conception of Wally/Waldo.

2nd one: Corona Version

Italian artist Pedro Mezzini has the answer: “Where is Wally, Corona edition”.

Found!

***

And what have you found recently?