A bicycle-share depot can label only 35 bent-wheel photos this month. Leadership still wants a junior mechanic to see a suggested mark and then accept or override it. They do not want a from-scratch model that needs thousands of new labels. Which approach fits?
Select an answer to reveal the explanation.
Short Explanation
Think of a junior mechanic with a suggested sticky note they can peel off. Thirty-five labels is not enough to train from scratch, so use a pretrained helper and keep the human override. An unreviewed score fires the mechanic, and a language model does not magically label wheel photos.
Full Explanation
Scarce labels and a required human override point to a human-in-the-loop assistant or a pretrained vision service, not a green-field supervised project. Waiting for thousands of new labels blocks a job that only has 35 tags this month. An unreviewed score replaces the decision leadership wanted a person to keep. A language model still needs the right kind of data and does not remove the label problem on photos.