A manufacturer needs to recognize product defects that are specific to its own items and has labelled example images. Which approach should it investigate?

1. Assume a general pretrained model already knows every proprietary defect category
2. Train an appropriate custom Vision image-analysis model
3. Rename each file to the expected defect and treat that as evaluation
4. Use only speech recognition on the image files