avatarDariusz Gross #DATAsculptor

Summary

This context discusses the use of Generative Adversarial Networks (GANs) to create an autonomous artistic immortality for a late sculptor, Siegfried Gross, by training an AI model to create art pieces in his style.

Abstract

The article explores the potential of GANs in preserving the artistic legacy of Siegfried Gross, a late sculptor, by creating an autonomous artistic immortality for him. The author, Dariusz Gross, aims to create an AI model, Artificial Sculptor Siegfried Gross (ASSG), using GANs and reinforcement learning. The training data for this model will include 3D scans, photogrammetry, point clouds, photos, and videos of Siegfried Gross’ works and sculpting techniques. The goal is for the AI model to create new sculptures, generate images of them, and publish them on social media independently. The article also discusses four potential use cases of GANs in the project's lifespan, including arbitrary style transfer, PoseNET, DeOldify, and 3D Dense Face Alignment.

Bullet points

  • The article discusses the use of GANs to create an autonomous artistic immortality for Siegfried Gross, a late sculptor.
  • The author aims to create an AI model, Artificial Sculptor Siegfried Gross (ASSG), using GANs and reinforcement learning.
  • The training data for the AI model will include 3D scans, photogrammetry, point clouds, photos, and videos of Siegfried Gross’ works and sculpting techniques.
  • The goal is for the AI model to create new sculptures, generate images of them, and publish them on social media independently.
  • The article discusses four potential use cases of GANs in the project's lifespan: arbitrary style transfer, PoseNET, DeOldify, and 3D Dense Face Alignment.

AI : “I look forward to working with you in the future”

Is AI capable of human creativity ? (code examples 2021)

🔵 GET FULL MEDIUM ACCESS

Multi-Level Pixel-Aligned Implicit Function for High-Resolution 3D Human Digitization DATAsculptor 2020 ©

a lecture summary 28th September, 2019, inauguration of AIArtists.org

A short story about creativity

Creativity has always seemed to be inextricably linked with being human.

The term “creativity” which takes its roots in the Latin “creo, creare” has evolved along with the development of humankind. The seemingly simple phenomenon of creativity is in fact a highly complex process. Contrary to many other phenomena, science doesn’t recognize any universal, authoritative definitions of creativity.

Ancient Europe considered art mainly as an act of creation from nothingness, in their own words, “creatio ex nihilo”.

Greek philosophers like Plato completely rejected the concept of creativity, preferring to see art as a form of discovery.

Throughout the last two millennia the realms of creativity and art coexisted in the same general space. In early modernity the criteria for something being a work of art broadened significantly as artists reached for newer and newer tools in their search for a fitting means of expression.

In the last half a millenium we have had 3 main schools of creative expression:

1. Divine domain

2. Artistic venues

3. Innovative activities

Creativity has always carried with itself this certain divine element, and today we face the historic challenge of granting machine learning that final distinct trait ascribed to higher beings.

The all-enveloping complexity of modern life demands that artists now grasp for a new tool: machine learning.

Thanks to machine learning, artists are now able to create hyperobjects, groups of individual objects that transcend our cognitive and perceptive abilities. Comparing an object to a hyperobject is like visualising one ant and trying to imagine all ants on the entire planet: a quantity so massive that our minds cannot comprehend it.

GAN — the machine learning artist’s tool

The original GAN was created by Ian J. Goodfellow who described the GAN architecture in a paper published in mid-2014. A GAN consists of two neural networks playing a game with each other. The discriminator (right) tries to determine whether information is real or fake, while the generator (left) tries to create data that the discriminator thinks is real.

How artificial intelligence can help us understand human creativity?myFATHERintheCloud.ai

A few months ago I started a project — myFATHERintheCloud.ai — which is dedicated to the memory of my father, Siegfried Gross.

Siegfried Gross 2018 DATAsculptor ©

The photo here has been taken on the 13th of February 2018, a day before a his major surgery.

The final goal of the myFATHERintheCloud.ai project is to give my father autonomous artistic immortality, so he can keep on creating without anybody’s help.

This goal requires the creation of Artificial Sculptor Siegfried Gross (ASSG), an AI model based on GANs and reinforcement learning which will create art pieces as if my father created them.

The training data include 3D scans, photogrammetry, point clouds, photos, and videos of Siegfried Gross’ works and sculpting technique. This assumes that the AI model will autonomously create new sculptures, create images of them, and publish them in social media posts.

The first milestone of the project is the carving out of a new sculpture by an industrial robot, which can be recognisably in the style of my late father’s works. This is what autonomous artistic immortality will look like.

Four potential GAN use cases in the project’s lifespan

1.Arbitrary Style Transfer in Real-time with Adaptive Instance Normalization.

The method involves an adaptive instance normalization (AdaIN) layer that matches the mean and variance of the content image with those of the style image. This state-of-the-art method achieves speed comparable to the fastest existing approach, but without the restriction to a pre-defined set of styles. Moreover, the approach allows the user to input variables and alter the output with content-style trade-off, style interpolation, colour & spatial controls, all using a single feed-forward neural network.

2. PoseNET: Performing real-time skeletal tracking of one or more people.

Pose estimation refers to computer vision techniques that detect human figures in images and video, so that one could determine, for example, where someone’s elbow shows up in an image. To be clear, this technology is not recognizing who is in an image — there is no personal identifiable information associated to pose detection. The algorithm is simply estimating where key body joints are. PoseNet can be used to estimate either a single pose or multiple poses, meaning there is a version of the algorithm that can detect only one person in an image/video and one version that can detect multiple persons in an image/video. The model is split into two separate versions because while the single person pose detector is faster and simpler, it is incapable of processing video feeds with multiple actors.

3.DeOldify

The point of this project is to colorise and restore old images and film footage.

There are now three models to choose from in DeOldify. Each of these has key strengths and weaknesses, and so have different use cases. “Video” is for video of course. “Stable” and “artistic” are both for images, and sometimes one will do images better than the other.

More details:

Artistic: This model achieves the highest quality results in image coloration, in terms of interesting details and vibrance. The most notable drawback however is that it’s a bit of a pain to fiddle around with to get the best results (you have to adjust the rendering resolution or render_factor to achieve this). Additionally, the model does not do as well as stable in a few key common scenarios- nature scenes and portraits.

Stable: This model achieves the best results with landscapes and portraits. Notably, it produces less “zombies”- where faces or limbs stay gray rather than being colored in properly. It generally has less weird miscolorations than artistic, but it’s also less colorful in general. This model uses a resnet101 backbone on a UNet with an emphasis on width of layers on the decoder side.

Video: This model is optimized for smooth, consistent and flicker-free video. This would definitely be the least colorful of the three models, but it’s honestly not too far off from “stable”. The model is the same as “stable” in terms of architecture, but differs in training.

4. 3D Dense Face Alignment

3DDFA allows for 3D face alignment. A dense 3D face model is fitted to the image via convolutional neutral network. The output of this model is an image of a 3D depth estimation on the input face.

This bust, an early sculpture by my father, was stolen while he was still in education, leaving this photo as its only remains. Using 3DDFA, I attempted a reconstruction of the sculpture, and its face in particular. In the first experiment, (see photo below) the GAN is still trained on regular, realistic shapes and proportions. It may not bring the most desirable results, as the sculpture is quite expressive and lacks the colour cues and contrast of human skin, hair and eyes. The GAN’s output is as follows:

Closing thoughts

My project is only beginning to gather momentum. The model is still early in its training to acquire the skills that took my father 60 years to learn.

I have worked with reinforcement learning for a long time and seen how allows the agent to find the most effective way to reach its goal. Sometimes its solutions may seem peculiar, but it all works towards the end goal.

With every step in my project I make progress, and just like the agent/walker in the image above, using a hybrid solution and incorporating various ML models, I will reach my goal and grant my father artistic immortality so that he can live on independently, create on without anyone’s help.

Look how many steps the walker is taking, I am prepared to take as many as it is.

I am still only getting started, but I think that, in a similar fashion, human creativity is only beginning to take off.

And this answers the title question:

We are currently only at the beginning of the human creativity.

Tools like my father’s chisel allowed man to create something out of nothing.

Now we are handling a tool that expands our creative options into infinity.

We are witnesses to a life-changing step in the history of mankind, where each of us — no exceptions — can finally achieve our individual Autonomus Artistic Immortality.

if you use or intend to use AI in your creations, please leave a comment

code examples:

  1. style_transfer

2. De_Oldify

a few thoughts, with which I deal in my work in April / May 2021 (list is constantly updated)

1. Human creativity is any innovation that is new and has never been thought of by others before.

2. Artificial Intelligence (AI) artwork can be found in a variety of ways, such as videos, software, or even social media.

3. neural networks are able to think like humans but not in the same way as humans because neural networks have a mechanism to learn things and then a new connection was created between neurons that allows them to learn faster than us humans working with connections we have already built in our brains called synapses. Neural networks have given artificial intelligence machines the ability to think similarly to us humans, but with more speed and accuracy than we could ever achieve because they can learn much faster

4.Origin, History and Usage of Words : The Latin term creatio means “creation”: its derivational suffixes also come from Latin. The word create was first used in English in the 14th century. It is found in Chaucer’s The Parson’s Tale, which is a story about divine creation. Time to expand this definition, please post a comment on this article, suggest a new definition

5. What is Artificial Intelligence (AI) artwork?

6. how human creativity came into existence ?

7. where does creativity come from?

8. Human creativity, artificial intelligence, neural network, machine learning, creativity, artist ,intelligence, , network, thinking, creative thinking, network, origin of creativity, neural network, adversarial network in XXI century, human cognition, human artist, divergent thinking, gene, data

9. evolution of genetic networks for creativity, creativity remains a frustrating puzzle. It is unclear why modern humans are so creative when our genetic codes are very similar to those of Neanderthals and chimpanzees.

10 .95% of the 267 genes we found only in modern humans were not protein-coding, including many long-non-coding RNAs in the self-awareness network. (nature.com)

11.What is creative AI (creativity, processes, portrait, visual artist, network, creative practice, creative potential, Theories of creativity, creativity skills. Generative Adversarial Network, model of creativity)

12. What is a Generative Adversarial Network (GAN)?

13. How can AI practitioners and artists work together?

14. What is deep learning?

thank you for reading, you may also enjoy some other articles:

All images created by the author.

Artificial Intelligence
Machine Learning
Art
Sculpture
Creativity
Recommended from ReadMedium