Transfer Learning with MobileNetV2 Accuracy Problem

I ran through the Transfer Learning with MobileNetV2 lab, and I was able to submit the assignment with full credit.

I think I may have gotten something wrong though, since the accuracy was fairly low, and fine-tuning seems to have had no impact on the accuracy:

image

Anyone know if this is expected? If not - where should I look to try to fix the problem?

I figured it out - I was not setting the initial weights correctly. It might be worth adding a validation on either the accuracy or the weight selection, if possible.