"Roxy AI" dramatically evolves appearance inspection with image generation AI!

Roxy AI's proud defect detection AI can accurately detect even very small defects and has contributed to the automation of inspections for many customers. However, the defect detection AI struggles with detecting defects that require a wide view of the image and unknown defects, which may not meet your expectations. The newly added good product learning AI excels at determining defects that the defect detection AI struggles with. Since the good product learning AI only learns from good product data, it can be used in environments where defects do not occur. There are three AIs that support the defect detection AI, including the good product learning AI: - Good product learning AI: Detects a wide range of defects, unknown defects, and human errors - Image generation AI: Generates high-quality training data for the defect detection AI - Zero learning measurement AI: Achieves size measurement and count without learning By combining these, flexible and high-precision inspection automation can be realized. A one-month trial is also available. Please give it a try.

Learning data is essential for AI training. The accuracy of AI will definitely improve if we can collect more diverse learning data. However, it is difficult to collect many samples for products that rarely have defects. Roxy AI's image generation AI can generate a variety of images as follows, which can reduce the challenges of sample collection: - Generate defective images by specifying the shape and type of defect - Specify the location where defects occur - Modify the shape of existing defects - Reuse an image generation AI trained on one product to generate defective images for another product (it can also generate defective images for products that have not yet started production and have no defects) Of course, normal images can also be generated. By generating and training similar images in areas prone to over-detection, we can reduce false positives. Training the image generation AI requires 10 to several dozen pieces of learning data. However, since we can learn from defects that occurred in different products together, the number of defective samples that need to be collected for one product is not that high. We can generate images that are difficult to distinguish from real ones. Please take a look at the sample images.