NIA·People and Forest, Cultural Assets·Products·City Landmarks Data Opening
Expected to be utilized in smart tourism, smart education, and smart store fields

The National Information Society Agency (NIA) will release 3.6 million large-scale Korean object images for artificial intelligence learning starting in January.
The images being released this time are ▲2.6 million images of nationally designated cultural properties such as palaces, pagodas, and castles, ▲800,000 images of products such as shoes, bags, and hats, ▲200,000 images of landmarks from 35 cities, for a total of 3.6 million images.

Considering that the competitiveness of artificial intelligence comes from securing data, this project was carried out with the goal of building a large-scale Korean-style object image for artificial intelligence learning.

In the field of artificial intelligence learning data, there has been a lack of datasets on object images that are suitable for the Korean situation.

Accordingly, the Korea Information Society Agency has been implementing a project to build and expand data for artificial intelligence learning since 2017, and started with building four types of datasets, including law, patents, general knowledge, and Korean images, and opened data on seven types, including tourism, agriculture, and healthcare, in 2019.

In addition, in May of last year, Media Group People and Forest, a company specializing in image DB construction, was selected as the business operator to build a Korean-style object image database and will work on it for 6 months. We have created over 3.6 million Korean-style object images and 3,000 images of each type.
The historic buildings designated as national cultural properties were designed by borrowing the Cultural Heritage Administration's classification system, and approximately 850 objects were photographed and collected, mainly in the Seoul, Gyeonggi, Incheon, and Gyeongju regions.

The products were collected based on the classification system of large online markets and about 250 objects, including precious metals, shoes, and cosmetics, were collected based on products preferred by foreign tourists and products produced and distributed in Korea. A total of 68 landmarks were selected in 35 cities across the country, including Jeju Island.

In particular, ADAM KB, which corresponds to the ontology dictionary of this project, has 26 million instances in 7 domain areas including people and places, and the number of triples after inference is 450 million, the largest in Asia.

An official from the Korea Information Society Agency said, “The 3.6 million Korean-style object image data sets that have been built this time can be viewed through the AI Hub website operated by the Korea Information Society Agency,” and “We expect that they will be utilized in various fields such as smart tourism, smart education, smart factories, and smart stores in the future.”