Removing the Black Box: The Inspection AI 'RoxyAI' That You Can See and Touch

basic information

**Features of RoxyAI** 1. Visualization of AI Personality (Perception) The performance of AI cannot be measured by numbers alone. RoxyAI visualizes the personality of the AI. 2. Understanding AI's Perception (Palpation) By tracing the areas of missed detections or over-detections with a mouse, you can confirm how the AI perceives them. When a defect is detected, the mouse vibrates (when using a compatible mouse), allowing you to gauge the AI's confidence level by the intensity of the vibration. 3. One-Click Annotation for Defective Areas You can simply click on the defective area, even if it slightly overlaps, and the AI will recognize it. In typical AI visual inspection software, it is necessary to indicate the defective areas accurately. 4. No Need for AI Engineers Anyone capable of visual inspection can easily create AI in a short time. Instead of focusing on parameter tuning, the process emphasizes improving accuracy, allowing for easy and instant adjustments to precision even after operations have begun. 5. Learning AI with Normal Products and a Few Defective Ones The required sample size is about 100 normal products and approximately 30 defective products per type. Additionally, it can detect unknown defects as anomalies even if they have not been previously learned.

Price information

The three machines are proposal-based and support our customers' "troubles." 【Purchase Method】 In addition to cash lump-sum purchases, we can propose payment plans that meet our customers' needs in collaboration with partner leasing companies. Recently, there are also leases that can ultimately be converted into company assets. Additionally, we can assist with various subsidies. If you have any concerns, please feel free to contact us.

Delivery Time

Applications/Examples of results

【Other Features】 - The AI automatically extracts suspicious data from the training data. - Focused checks on intervals where normal and defective data are mixed on the histogram. - Human judgment is used to handle suspicious data by reviewing enlarged images and overall images. - If there are issues, the corresponding images can be deleted or the normal/defective labels can be swapped. *This does not guarantee the detection of all defects.