The learning data generation function is a function that uses generative AI to create data that can replace high-quality data that is difficult to obtain in actual industrial fields such as medical, manufacturing, construction, and distribution to create vision AI models. It was developed to secure rare data such as defective products, natural disasters, and accidents that do not easily occur in the field.
For example, in order to develop an AI model that detects fire hazards in a manufacturing plant, it is necessary to sufficiently learn images or videos of actual fire situations, but since it is a special situation that does not occur frequently, there are difficulties in securing data. In addition, the occurrence rate of defective products in a typical manufacturing line is less than 10,000 to 1, so it is not easy to secure image or video data of situations where defective products occur.
Superb AI has added the ReCo generative AI model, which has been pre-trained with various data, to its self-developed and operated vision AI all-in-one platform, the Superb Platform. If this model is additionally trained with a small amount of data of less than 100 sheets, high-quality training data can be generated immediately. ReCo is an image-generating AI model developed by Microsoft, and is known to output object locations within images more precisely than existing image-generating models. It plans to continuously add the latest models that can perform better in high performance or other types of generation.
When you select the data and quantity you want to generate, the generative AI model automatically generates learning data, so you can immediately use it for model training. You can generate data in more detail, such as by generating the entire image, synthesizing only the surrounding environment while fixing the structure in the image, and synthesizing the detailed shape of the structure while fixing the background. It also provides a function to input a command to more specifically set the characteristics of the data you want to generate.
Meanwhile, Superb AI is operating the ‘Superb Service’, which supports the entire process from consulting to data design, AI algorithm development, and operation by experts, and the ‘Superb Platform’, a high-performance AI construction platform that covers the entire AI development cycle on a single platform. Since data analysis, model production, and distribution are possible within a single SaaS platform, AI models can be developed and managed more quickly.
Superb AI CEO Hyunsoo Kim said, “As the field changes and industries become more advanced, the product defects and specific situations that AI models must recognize are also becoming more diverse,” and “By utilizing the learning data generation function, we will be able to supplement the insufficient data and increase the range of responses to model performance.”
editor@itworld.co.kr
Source: www.itworld.co.kr
