A bakery's Custom Vision model recognizes its current cake boxes. The company introduces a new seasonal box design, and the model misses it entirely. What should the team do?
Show answer and explanation
Answer: B
Custom Vision classifiers only learn from images that have been labeled and included in training. A new design represents a visual class the model has never seen, so the team must add tagged examples and train a new iteration. Adjusting thresholds only changes how existing predictions are accepted, and it cannot teach the model a pattern it has never been shown.
Why the other options are wrong
- A. Changing the confidence threshold alters which predictions are reported, not what the model has learned. The new box would still go unrecognized, and a higher cutoff would likely make the misses worse.
- C. SSML controls how text is spoken by a speech service and has no role in image recognition. It cannot teach a vision model anything about packaging.
- D. A search indexer builds a queryable index from documents and does not update a classifier. Indexing catalog pages would not change what the image model can recognize.