• Derek Philip Au
  • Work
  • About
  • Contact

Derek Philip Au

Derek Philip Au

  • Derek Philip Au
  • Work
  • About
  • Contact

Impasto Dishes

by Derek
July 10, 2020

This was just a first try, didn’t come out like I was hoping.

Categories Ceramics

Post navigation

Previous Post Last Kiln Before QuarantinePrevious Post
Next Post Tsang Tsou Choi 曾灶財 King of KowloonNext Post

Categories

  • Life (12)
    • Family (10)
      • Heirlooms (7)
  • Photos (95)
    • Jingdezhen (79)
  • Uncategorized (3)
  • Work (148)
    • Ceramics (117)
      • Antiques (6)
      • Exhibitions (3)
      • Glazes (73)
        • Porcelain Body (1)
        • Tests (15)
        • Traditional (12)
        • Underglaze (1)
      • Technique (29)
    • Software (24)
      • Glazy (16)
      • Machine Learning (4)

Follow Derek on Instagram

I'm a huge fan of @glaze_gumpu's work and his expl I'm a huge fan of @glaze_gumpu's work and his exploration of the fundamental properties of glaze- color, fluidity, transparency, & surface.  And for the glaze nerds out there, check out that scale!  https://gumpunakamura.myportfolio.com

https://www.youtube.com/watch?v=da0_ohNUrPk

#craft #glaze #glazes #glazerecipes #glazetest #ceramics #pottery #ceramicglazes #glazechemistry #glazechem #glazecalc #umf #glazy
Just some of the photos uploaded this week! Durin Just some of the photos uploaded this week!

During the first couple years of Glazy's existence, it would be exciting to see just one or two photos uploaded a week.  Since then, the Glazy community has grown into an amazingly generous group of people from around the world, sharing over 30,000 photos!

All these photos with recipes are available on https://glazy.org (link in bio)

Photo credits:
Wiliam Kaplan
Barbara Nowak
serene fang
Duncan Mackellar
Alice Salmon
Tara
Wilson
Christine Bouma
Donte of earth nation
Jeffrey Yauger
Tom Demeranville
Ginger Lukas
Kaitlin Devine
Sydney Baker
Randall Johns
Yvonne
HP Bloomer IV
Matt Kelleher
Jake Glaze
Bill Kitt
Michaela Danzl
John Post
Martins
Ruau Francoise
Jorge H Morales
Jun Yu Tan
Atelier Aimee
Martin Hill
Rylie McGuire

#craft #glaze #glazes #glazerecipes #glazetest #ceramics #pottery #ceramicglazes #glazechemistry #glazechem #glazecalc #umf #glazy
Hope for 2021. Joseon Porcelain with gold lacquer Hope for 2021.

Joseon Porcelain with gold lacquer repairs, National Museum of Asian Art.

https://asia.si.edu/object/F1897.87/

#kintsugi #mend
I asked OpenAI's GPT-3 to generate haikus about po I asked OpenAI's GPT-3 to generate haikus about pottery.

#generatedart #openai #gpt3 #machinelearning #generative #generatedceramics #craft #ceramics #pottery #art #glaze #glazes #glazerecipes #glazy
This inspirational art quote does not exist. Prom This inspirational art quote does not exist.

Prompted with a list of real quotes by famous artists, OpenAI's GPT-3 was asked to generate a list of fictional quotes by fictional artists.  A small percentage of responses were plagiarisms, while others were nonsensical or trite cliches.

Presented here are curated examples paired with StyleGAN generated art from thisartworkdoesnotexist.com.

Some of these quotes are partial plagiarisms or based on patterns found in real quotes.  For example "Art is the objectification of feeling" is attributed to Herman Melville, while "Art is the objectification of feeling, and the subjectification of nature" is attributed to Susanne Langer.  GPT-3 came up with its own version:  "Art is the objectification of imagination seen through the eyes of empathy."

#generatedart #openai #gpt3 #machinelearning #generative #generatedceramics #craft #ceramics #pottery #art #glaze #glazes #glazerecipes #glazy
This vessel does not exist:  Vase + dress hybrid This vessel does not exist:  Vase + dress hybrids.

Using a StyleGAN2 model trained on 40,000 images of vases, transfer learning with a dataset of images of dresses.

#thisvesseldoesnotexist #stylegan2 #stylegan #gan #machinelearning #generative #generatedceramics #vases #ceramicfashion #ceramics #pottery
This glaze did not exist. As detailed in my previ This glaze did not exist.

As detailed in my previous post, I used 10,000+ public Glazy recipes to fine-tune the GPT-2 language model.  This training allowed me to generate thousands of "fake" glaze recipes.  Surprisingly, most of these recipes looked like they would work.  The obvious next step was to fire some real tests of these "fakes".  Rather than make a random selection, I chose recipes that I thought would fire to maturity in cone six oxidation.

The results were quite good, and I posted two of the recipes on Glazy.  There are hundreds of other fake recipes for you to try!  (Link in bio.)

#openai #gpt2 #machinelearning #generative #generatedceramics #craft #ceramics #pottery #art #glaze #glazes #glazerecipes #glazy
This glaze recipe does not exist. Released early This glaze recipe does not exist.

Released early last year, GPT-2 is a large language model with 1.5 billion parameters trained on a dataset of 8 million web pages.  The model can be "fine-tuned" with custom data sets, for example collections of poetry or cooking recipes.  After training, these custom tuned models can generate "fake" texts, e.g. poetry or recipes.

I fine-tuned the GPT-2 355M model using over 10,000 public Glazy recipes as training data. Although the loss was still going down after 6,000 steps, I stopped training at that point as I noticed more duplicates being generated and I was afraid of overfitting my relatively small dataset.  I then generated a couple thousand "fake" recipes using two different "temperature" settings, which controls the "creativity" of the output.

Example "fake" recipe:

Amber Celadon
34.0 Albany slip
20.0 Custer Feldspar
13.0 Silica
13.0 Wollastonite
6.0 Whiting
3.0 EP Kaolin
3.0 Gerstley Borate
3.0 Rutile
2.0 Red Iron Oxide

The results for both sets were surprisingly good: At first glance the recipes seemed "real" with proportional mixes of feldspars, clays, silica, fluxes and colorants/opacifiers. Even the total ingredient amounts added up to a reasonable number, usually in the range of 90-110%. The duplication rate was about 5% for t=0.7 and 4.5% for t=0.9.

After removing duplicates, I imported these fake recipes into Glazy in order to see their resulting analyses and visualize them on the Stull Chart. I was surprised to find that, as with "real" glaze recipes, most of these generated recipes fell comfortably within the major Stull regions of Bright, Matte, and Semi-Matte. The set with temperature setting of 0.9 seemed to generate more variation.

This weekend I am firing some of these fake recipes and will post the results soon!

If you are interested in trying these fake recipes for yourself, the files can be downloaded from:
https://thisvesseldoesnotexist.com/#/about (link in bio)

And yes, I'm still waiting for access to OpenAI's GPT-2 successor, GPT-3 :)

#openai #gpt2 #gpt-2 #machinelearning #generative #generatedceramics #craft #ceramics #pottery #art #glaze #glazes #glazerecipes #
Load More... Follow on Instagram

Derek Philip Au

Menu

  • Derek Philip Au
  • Work
  • About
  • Contact
  • Copyright © 2019 Derek Philip Au