What is the main advantage of using transfer learning in computer vision?

Correct answer: It requires less training data and reduces training time

Option A is correct because transfer learning leverages weights pretrained on a large dataset, allowing a model to achieve strong performance on a new, related task with far less labeled training data and significantly shorter training time compared to training from scratch. Option B is wrong because transfer learning does not remove the need for feature extraction; the pretrained layers have already learned meaningful feature representations, but those representations still perform extraction, and fine-tuning stages may still involve feature-level adjustments. Option C is wrong because transfer learning does not eliminate the need for labeled data; labeled examples of the target task are still required for fine-tuning, just fewer of them. Option D is wrong because no machine learning technique guarantees perfect accuracy; performance depends on data quality, domain similarity, and model design.