Week 9 - OpenAI Scholars Final Project Proposal and Dataset Attributes - Generative Art with Domain Knowledge

Submitted by hollygrimm on Mon, 08/06/2018 - 09:08


As a painter and software developer interested in Machine Learning, I have trained several models on my paintings, creating complex and interesting artistic outputs from CycleGAN and Mixture Density Networks (modeling distributions of data). For my final project, I would like to expand on my generative art projects by incorporating domain knowledge based on the rules of art evaluation.

Generative Adversarial Networks (GAN) with Multiple Discriminators

I will be using a machine learning algorithm, like CycleGAN, with multiple discriminators, to generate unique works of art. Each discriminator will consider some of the following art attributes:

  • Technical Merit
  • Composition
  • Color
  • Style

I learned this method of evaluating artworks from my fantastic art teacher in high school, Sherry Rosevear. She had us memorize three acronyms: DATE, ACT, and VCRUB. DATE means to Describe, Analyze, Translate, and Evaluate. Analyze is broken into Composition and Technical Merit. Finally, Composition includes Variety, Contrast, Repetition, Unity, and Balance. I will go into more detail about these terms below.


The GAN dataset will be composed of my paintings and drawings.

For the specific discriminators, I will be training on Kaggle’s WikiArt "Painter by Numbers" dataset. All the images are protected by copyright and can be utilized only for the purposes of data mining, which constitutes a form of fair use. This week I started to label each of the WikiArt images with art attributes. I need to complete this project in four weeks by the end of August 2018, so I’ll only be able to label a subset of the 100,000 images in the dataset.

The “Style” attribute has already been captured in the WikiArt dataset, with each artwork labeled with one of 27 styles:

  • Abstract_Expressionism
  • Action_painting
  • Analytical_Cubism
  • Art_Nouveau
  • Baroque
  • Color_Field_Painting
  • Contemporary_Realism
  • Cubism
  • Early_Renaissance
  • Expressionism
  • Fauvism
  • High_Renaissance
  • Impressionism
  • Mannerism_Late_Renaissance
  • Minimalism
  • Naive_Art_Primitivism
  • New_Realism
  • Northern_Renaissance
  • Pointillism
  • Pop_Art
  • Post_Impressionism
  • Realism
  • Rococo
  • Romanticism
  • Symbolism
  • Synthetic_Cubism
  • Ukiyo_e

Technical merit ranges from 1-10. In my opinion, most of the artworks in the WikiArt dataset have a technical merit of 10. Unity, or the elements that force you to look at the center of interest, has also been 10 for all the artworks thus far. Finally, I found most of them to be Balanced. Since the WikiArt dataset doesn’t have a wide range of values for these attributes, I’ll have to wait until I have a different dataset to train discriminators on these features.


Attributes for Training

Variety of Texture (values 1-10)

Barnett Newman’s Who’s Afraid of Red, Yellow and Blue II has very little texture:

Barnett Newman Who’s Afraid of Red, Yellow and Blue II

Whereas Konstantin Korovin’s Paris Seine has a lot of texture:

 Konstantin Korovin Paris.Seine


Variety of Shape (values 1-10)

Lucio Fontana’s Concept Spatiale has only a few shapes:

Lucio Fontana Concept Spatiale

Jean Fouquet’s Funerals has a variety of different shapes:

Jean Fouquet Funerals


Variety of Size (value 1-10)

Patrick Henry Bruce’s Composition I has shapes that are all about same size:

Patrick Henry Bruce Composition I

Volodymyr Orlovsky’s Reaping. Hiok. has many different sizes of shapes:

 Volodymyr Orlovsky Reaping. Hiok


Contrast (value 1-10)

Tosa Mitsuoki’s Night March of a Hundred Demons (left half) has low contrast:

Tosa Mitsuoki Night March of a Hundred Demons (left half)

Samuel Peploe’s Still Life with Roses in a Chinese Vase has high contrast:

Samuel Peploe Still Life with Roses in a Chinese Vase


Repetition (value 1-10)

Gene Davis’ Homage to Dubuffet I has only very little repetition of shape:

Gene Davis Homage to Dubuffet I

Josef Albers’ Frontal has a lot of repetition of shape:

Josef Albers Frontal



The primary color is selected using the Magenta, Cyan, and Yellow color chart which includes the following values: Magenta, Red-Magenta, Red, Orange, Yellow, Yellow-Green, Green, Green-Cyan, Cyan, Blue-Cyan, Blue, and Blue-Magenta.

The harmony feature can be one of six: monochromatic, analogous, complementary, split complementary, triadic, tetradic.

Hiroshige’s Small Bird on a Branch of Kaidozakura is Orange/Split Complementary:

Hiroshige Small Bird on a Branch of Kaidozakura

Franz Richard Unterberger's Procession in Naples is Orange/Complementary:

Franz Richard Unterberger's Procession in Naples

Francisco Goya’s Bildzyklus is Orange/Analogous

 Francisco Goya Bildzyklus

Josef Albers’ Frontal is Orange/Monochromatic:

 Josef Albers Frontal


Next Week

I’ll continue to label the WikiArt dataset and start training my first discriminator.