Defective Image Generative AI
What is Defective Image Generative AI?
Defective Image Generative AI is an innovative AI that learns from product image data to generate a large number of good/defective product images. With the generated good/defective product images, users can easily create a large amount of data for inspection systems. The accuracy of automated visual inspection can be significantly improved with the generated images.
One of the major challenges for manufacturers when introducing automated visual inspection is the shortage of defective product images, which are essential for algorithm training. While inspections can be performed with good product images only, it is desirable to use defective product images to improve system accuracy.
To solve this issue of "lack of defective product images", RUTILEA has developed a technology that generates a large number of good/defective product images from a small number of product images, using a generative model based on deep learning. The ability to generate high-resolution images using diffusion models is a feature of this technology.
By utilizing about 10 defective product image data, a large number of defective product images can be generated, which can be used to improve the accuracy of the visual inspection system.
Why Choose Rutilea's Defective Image Generative AI?
① Reducing the hassle of collecting training data
Usually, collecting good product images and defective product images requires a lot of manpower and time, but our product can automatically generate images, reducing the burden on workers, realizing efficient data collection, and improving project progress speed.
② Resolving the shortage of defective product images
Even if the number of defective products is small, our product can generate a large number of defective product images, which can smoothly advance the automation of visual inspections.
③ Improve inspection accuracy through high-quality images
Good/defective product images can be generated in high quality and in large quantities. In a verification example, the inspection accuracy improved up to 97%.
An achievement of 97% inspection accuracy with approximately 900 defective images