Tool Operation Guidelines
| Setting | Maximum supported value |
| Maximum image size | Images up to 32MP (up to 16k in height or width) are supported |
| Maximum number of images per stream | Red Analyze tool = 1000 Blue Locate, Blue Read and Green Classify tools = 25000 |
| Blue Locate and Blue Read tools Maximum number of trained features in an image Maximum number of discrete labeled features |
1000 256 |
| Red Analyze Maximum number of defects per image |
1000 |
| Green Classify tool Maximum number of Focused Green Classes Maximum number of HDM Green Classes |
1000 200 |
Application Development Tips
- Clone the tool and perform AB testing on it.
By cloning a tool, you can change the parameters and re-train them, compare the original to the clone, and see the effect of the tool parameter changes. - Create a duplicate of the stream and test the image set.
If you have a lot of image processing that you want to use for testing, but you want to keep the database, duplicate the stream and add the image to the duplicate. You can then process and test the state of the current tool. When you are finished, delete the duplicated stream. - Don't add duplicate images to "blur" the tool.
This has a negative impact on the model. Alternatively, if you want to increase the weight of training a particular picture, use Class Weights option of Green Classify in High Detail. - Perform "bootstrap" labeling.
Bootstrap labeling refers to the process of starting the labeling process by simply labeling the small sample size of an image, then training the tool, accepting views with the correct markings and converting them to labels, removing incorrect markings, and checking the results. Replace them with the correct labels. You can further accelerate this process by splitting the image into multiple views and labeling only one view. In addition, you can temporarily reduce Epoch Count parameter (either). This will reduce your training time
Tool Tips
- To focus the tool on the center of the ROI, add a masking around the border of the view.
- If your application includes the Rotate feature, enable Oriented parameter in Blue.
- If your Unsupervised Mode uses Red Analyze, use View Inspector to set Iterations and Training Passes parameters.
