Create a New Model#
This page explains how to create a reference model for component recognition.
Step 1: Physical setup preparation#
1 |
Remove the calibration grid and restore the initial layout:
|
2 |
Position one object at the center of the viewing area |
Step 2: Access the model#
Once physical preparation is complete, proceed with image acquisition and model creation.
3 |
From the Recipes page, with the correct recipe selected, click Edit Recipe |
4 |
Select the FlexiBowl currently in use |
5 |
The available model slots will be shown, up to 8 models per recipe |
6 |
Click Model 1 to open the Train Model 1 Cam 1 page |
Train Model interface overview#

Parameter |
Function |
|---|---|
Enable Model |
Activates this model slot and makes it usable |
Grab Train Image |
Captures a photo of the reference component for training |
Score Threshold |
Adjusts the model detail level, from 0 for maximum detail to 1 for minimum detail |
Train |
Generates the model by processing the acquired image |
Model Name |
Text field used to assign a descriptive name to the model |
Tip
Managing multiple models
At this stage only the first model is activated. After completing it, it will be possible to:
Enable additional slots, Model 2, Model 3, and so on, for different parts inside the same recipe
Modify existing models
Disable models that are no longer required
For now, focus on completing the first model.
Step 3: Training procedure#
7 |
Click Enable Model to activate this model. The model is now active and ready to be configured. |
8 |
Click Grab Train Image to capture a photo of the reference component positioned on the FlexiBowl Warning The reference component must remain fixed in that position throughout the entire application creation process. |
9 |
Move the ROI box so that it fully frames the component |
10 |
Move the origin reference point to the center of the ROI area Tip Where should the origin be placed? The origin is automatically placed at the center of the component.
The origin defines point (0,0) of the model coordinate system. |
11 |
Use Score Threshold to adjust the desired detail level Note Score Threshold
Value close to 0 -> detects MORE details, resulting in a more precise model Value close to 1 -> detects FEWER details, resulting in a simpler model Tip How to choose the optimal Score Threshold Use a LOW value, 0.1-0.3, when:
Use a MEDIUM value, 0.4-0.6, when:
Use a VERY HIGH value, 0.7-0.9, when:
|
12 |
Click Train |
Step 4: Visual check#
After generating the model, it is essential to verify its quality before proceeding.
13 |
Zoom into the image to inspect the created model in detail and verify that it is correct Tip Characteristics of a valid model ✓ It contains enough lines to recognize the component ✓ It does not include the texture of the background surface ✓ It avoids light reflections
|
Attention
If the model is not satisfactory:
Modify the Score Threshold
Click Train again
Repeat until the model is optimal
Tip
Optimization strategy
Problem: model includes background surface texture
-> Solution: increase Score Threshold or increase the Cam Exposure value in SETUP > Camera Setup > Cam Exposure
Problem: model has too few lines and is not distinctive enough
-> Solution: decrease Score Threshold
Problem: model includes reflections
-> Solution: increase Score Threshold or adjust camera exposure
Apply gradual changes, typically 0.1 to 0.2 steps, and test every time.
Step 5: Saving#
14 |
Name the model with a descriptive name Tip Avoid generic names ❌ Names to avoid:
✓ Recommended names:
A clear name makes management easier when many different models are available. |
15 |
Click Next -> the Define Robot Pick Area page opens |
See also
Proceed to ROI Definition to continue configuration.

