How many training images are needed for AI visual inspection? How to determine and proceed.
The number of images required cannot be expressed as a single number. It is determined by the degree of variation in good products, the number of types of defects, and the stability of the shooting method. If the variation is small and the shooting method is stable, you can start with fewer images. When defects are not accumulating, you can begin with only good products and introduce a few defective items for verification purposes. 【Key Points】 ■ The number of images is determined by variation, types of defects, and the shooting method. ■ If the shooting method is unstable, no matter how many images you collect, it will not be sufficient. ■ When defects are not accumulating, you can start with good products. *For detailed explanations, please refer to the technical notes and downloadable materials in the related links.
Inquire About This Product
basic information
**Contents of this Technical Note** - The reason the number of images cannot be determined is that there are multiple factors to consider. - Before determining the number of images, check if the shooting method is consistent. - How to proceed when defective products do not accumulate. - Precautions when augmenting images. - Cases where increasing the number of images does not solve the problem. - How to verify if there are enough images after learning. - The required amount of images should be estimated after photographing the actual items. **Consultation** I will confirm whether it can be measured or captured with the actual item. Please consult me starting from one photo.
Price information
The technical notes can be read by anyone from the related links. We will discuss the costs of development and testing based on the details you provide.
Delivery Time
※It varies depending on the content. Please contact us.
Applications/Examples of results
【Frequently Asked Questions】 Q. Can I start even if there are no defective products? A. Yes, you can start if you are learning only with good products. However, you will need actual defective items to confirm whether you can detect the targeted defects. You will need to borrow samples like limit samples to test. Q. Is it better to have more images? A. Having more images does not necessarily mean better results. If you increase the number of images while the capturing method is shaky, the accuracy will be harder to improve. It is efficient to collect images that meet the conditions you need within the necessary range. Q. What should I do if new defects appear after training? A. You will need to add those images and retrain. You will determine what to add based on the actual images of the defects. It is easier if you have a system in place from the beginning to accumulate images during operation.
catalog（1）
Download All Catalogs![[Technical Document] Automation of Visual Inspection: Approach and Common Pitfalls, How to Capture Images and Choose Lighting Before AI](https://image.mono.ipros.com/public/catalog/image/2115800/IPROS9310882243773939791.png?w=120&h=170)
Company information
We specialize in 2D and 3D inspection and measurement technologies using cameras. We possess a wide range of knowledge in optics, mechanics, electronics, and software, enabling us to provide technical support for both hardware and software. Not only do we have personnel with experience working as researchers at image inspection device manufacturers, major electronic component manufacturers, and universities, but we also have numerous successful implementations of developed products in actual production environments, making our ability to adapt advanced technology to on-site needs a strong point. Additionally, we are skilled not only in developing proprietary algorithms but also in using off-the-shelf products such as image processing controllers from Keyence and Omron, allowing us to excel in selecting and utilizing the most suitable equipment for specific needs. In recent years, while AI technology has garnered attention, there are cases where utilizing 2D and 3D technologies can be implemented more easily than AI, and cases where optical devices and mechanical jigs can improve performance and reduce software load. We offer services related to image systems, including "equipment selection," "implementation support," "performance improvement of existing equipment," "technology development," and "joint development," so please feel free to consult with us first.
