Neucraft — Exploring traditional crafts using AI — Part I — Data Collection

In the following series of blog posts we will present Neucraft, a project lead by ITOCHU Techno-Solutions (CTC) in collaboration with Asahiyaki that Qosmo has helped develop, which brings together modern AI technologies and traditional crafts with centuries of tradition behind them. This series of blog posts is available in Japanese on the project’s webpage https://neucraft.ai/, and will be published in English on Qosmo Lab.

Part I. Looking back on 400 years of history — Data Collection

The beginning of NeuCraft

The Uji River in Kyoto near the Asahiyaki kiln

“Using AI to convey the appeal of traditional crafts to a wider audience” is the concept that started the first project of Neucraft, a collaboration with the Asahiyaki kiln in Uji, Kyoto, which has a tradition of over 400 years.

There are various quality factors in ceramics, such as shaping, glazing, texture, firing and we have had many discussions with Mr. Matsubayashi, the head of Asahiyaki, about where AI can contribute to the production process.

Experience is required throughout the production process, especially in the firing process, which requires a high degree of skill to control the process while at the same time being a creative process that produces unexpected results. In addition, Mr. Matsubayashi, who places importance on “making pottery that is well received by the people of the time” expressed his hope that AI would give him design ideas that he would not have thought of from his own perspective.

With the recent development of generative AI technologies such as GANs and Diffusion models, the development team of ITOCHU Techno-Solutions’s AI Business Department has been keenly interested in the use of this technology. They have been asking how AI can model the 400+ years old tradition of Asahi ware, provide us with new insights, and generate new artworks from it. And through this process, we would like to explore the possibility new value creation for ITOCHU Techno-Solutions’ customers as well.

Data collection as a historical retrospective

We set up a space just by the climbing kiln for a week-long photo shoot of Asahiyaki artworks.

In order to take on the challenge of a highly original project, we set out to collect our own training dataset with the cooperation of Asahiyaki, rather than collecting image data online, as is generally the case for AI training. The Asahiyaki kiln in Uji, Kyoto, houses thousands of historical artworks from 16 generations of potters.

Although it is true that the more photos required for study, the better, we found that a few hundred to a thousand photos can yield interesting results if the angles and lighting conditions are right, so we aimed to take 1,000 photos of the works from multiple angles over a five-day period.

We used a space beside the huge climbing kiln that has been passed down through the generations, set up the background and lighting, and used three cameras to shoot efficiently. However, when shooting from several angles at the same time, it is more difficult than expected to create optimal lighting at all angles. Even if the lighting was perfect from one angle, it was sometimes too bright or cast shadows when viewed from different camera angles, and it took longer than expected to set up the equipment, especially on the first day.

However, as the team became more adept with time, they gradually picked up the pace, and eventually completed their goal of photographing 1,000 works.

Here are some examples of the images that were taken and post-processed by the photographer for color adjustment. All of the works we photographed, both old and new, were very beautiful, and it was a valuable experience for us to be exposed to such precious works of art. Asahiyaki was also pleased that we were able to catalog their works once again through this photoshoot.

Checking the photographed data in the Python environment.

Using the photos obtained in this way as input, this first NeuCraft project will now proceed to build an original AI model using deep learning techniques. Since we were able to obtain a very high quality data set, we have high expectations for the results. Since such a project, applying latest technologies to traditional crafts is uncharted territory, we expect to discover many unanswered questions along the way, but this uncertainty makes the project even more exciting.

In this blog, we will continue to share with you the progress of the project and what we have learned from it at each step along the way.