Red Analyze Tool – Unsupervised Mode

When the Red Analyze tool is in Unsupervised Mode, the focus is on teaching the network what a good part looks like. During training, it is recommended to only present passing images of your part (including any and all acceptable variations). Then, during runtime, if a part is presented and the tool compares it against what it was taught and finds that it deviates, it will indicate that it thinks that the part is bad.

In Unsupervised mode, the process for training the tool is:

Collect images that represent the full range of passing and acceptable variations of your part.

Label the images as "good" by clicking each image in your Training Set.

A single click equals a "Good" image (denoted by a green stripe in the upper-right corner), and clicking twice equals a "Bad" image (denoted by a red stripe in the upper-right corner).

"Good" Label	"Bad" Label

Train the tool.

Validate the tool by presenting images that contain known defects to determine the tool's accuracy.

"Good" Marking	"Bad" Marking