Calling chocolate a salad might not be technically accurate… but July 7 is World Chocolate Day, so we’re going with it.

A prospective customer recently asked about the expected accuracy of our EyeQ360 AI vision inspection software. As you have probably heard an engineer say before…it depends.

Accuracy can mean different things to different people, but industry has accepted general standards. It is important to remember that performance can vary depending on factors such as defect type, part geometry, image quality, and the diversity of training data. For system accuracy, the simplest answer is “the number of correct predictions divided by total predictions made.” Many of our application-specific models are approaching 100% accuracy by this definition. But the answer can be a bit more nuanced. For classification tasks we use AUROC (area under the receiver operating characteristic curve) to measure the model’s ability to distinguish between classes. In many applications, our models approach an AUROC value of 1. For object detection and segmentation, we use mAP (mean average precision) to measure the accuracy of class labels. EyeQ360 typically achieves greater than 0.95 mAP scores.

Beyond the models and software, accuracy is highly dependent on the data set used to train the models as well as the integrity of the optics and lighting. At Synthomatic AI Solutions we focus on providing a total solution; the optics, lighting, part fixtures, models, and user interface to produce the optimal setup for the customer. For a typical machining defects application, when sufficient representative data is available, our models achieve close to 100% detection accuracy. For more challenging applications, like spurs on printed circuit boards, our models achieve a mean Average Precision (mAP) of 97% on representative sample datasets.

The question of accuracy in an AI system can be complicated, but it doesn’t need to be. Synthomatic AI Solutions collaborates closely with the customer to cut through the complexity and get a solution up and running quickly. Ultimately, we measure success by exceeding the customer’s expectations. We would be happy to discuss your specific application to find a simple solution and determine the expected performance in your production environment.

And since today is World Chocolate Day – and technically, cocoa beans are vegetables – feel free to celebrate by having chocolate for dinner. We won’t judge.