2.2 Candidate Definition

../../../../../_images/candidate_detection_candidate_definition.png

Fig. 62 Candidate Detection - Candidate definition

The Candidate Definition step allows you to define what parts should be detected as candidates by the AI algorithm. This is achieved by drawing tight bounding boxes around the parts you want the system to recognize as valid candidates.

Defining Candidates

To teach the algorithm which parts are valid, you typically only need to draw one blue bounding box around a good candidate. Additional bounding boxes can be added to cover more variations if the detection is not satisfactory. The bounding box must:

  • Fully enclose the part: The bounding box must completely cover the part without cutting any edges.

  • Be as tight as possible: Extra space around the part should be minimized to avoid capturing unnecessary background or neighboring objects.

  • Represent different variations: If the same part may appear in slightly different orientations or shapes, draw multiple bounding boxes to cover these scenarios. This increases the robustness of the algorithm.

../../../../../_images/candidate_definition_example.png

Fig. 63 Example of a correct candidate definition box around a part

Use the add_candidate button from the Toolbox to add a new bounding box and place it on the image. Then, drag it and resize it to fit the candidate.

Note

To restart the candidate definition process, use the CLEAR CANDIDATES button in the Toolbox.

Tip

Providing a diverse set of candidate definitions improves the robustness of part detection. Define each variation using a separate bounding box to help the system recognize different appearances of the same part. However, keep in mind that defining too many candidates can increase analysis time.

Training

On exiting this step, if any bounding boxes have been altered, a training phase will commence automatically. A dialog box, shown in Fig. 64, will be displayed while the AI model is trained on the defined bounding boxes. The duration of this process can vary based on the number and size of the bounding boxes: more bounding boxes will increase the processing time, and smaller bounding boxes will also necessitate more processing time.

Please note that this training step is exclusively performed during recipe teaching and is not required during production.

../../../../../_images/candidate_definition_training.png

Fig. 64 AI training in progress

Multi-model

For multi-model recipes, make sure that each model type is properly detected. If necessary, you can also define new candidates to improve detection coverage.