In version 3.7, we enhanced point annotations by allowing each label to be assigned an anomaly class. In version 3.8, criticality levels were introduced, making it possible to qualify each defect according to its impact: Critical, Major, Minor, or to mark it as an acceptable anomaly.
Version 3.9 fully leverages these criticality levels with two new features to go even further in the analysis and evaluation of sorting performance.
When adjusting sensitivity, you can now calculate calibration metrics according to the defects’ criticality level, using a drop-down list. The selection is multi-select: you can combine several levels at the same time to refine the analysis.
This filter is particularly useful to ensure, for example, that all critical defects are correctly detected by the system, by isolating that level in the metrics. By default, all criticality levels are included.
By refining your analysis of Spark’s performance by criticality level, you can optimize its calibration—and therefore the rejection rate—without compromising on the most important defects.
Precise and unambiguous annotation is key to ensuring a good understanding of quality criteria and optimizing Spark’s performance accordingly. That’s why we redesigned the manual evaluation to make it more complete and intuitive.
Users will be asked to point out the defects on the image when the part is defective. Beyond localization, the defect type and its criticality can also be filled in.
This new workflow is at the heart of a set of new features enabling better communication around part quality and improving analysis of production and sorting performance.
This feature is enabled by default on new projects. It will be rolled out more broadly in stages. Contact Scortex if you would like your kits to benefit from it.