Exercise: Fine-Tune Pretrained Classification CNN

Problem statement

In this lesson, we’ll fine-tune a pretrained MobileNetV3 classification CNN on the CIFAR-10 dataset. CIFAR-10 is a dataset of 32x32 color images from 10 categories (airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck). The training subset contains 50000 examples.

Note: The images are blurry because their size in the dataset is 32x32.

The classification task is much more difficult than the handwritten digits from the MNIST dataset. Therefore, it is wise to use fine-tuning on a pretrained classification CNN instead of training from scratch.

Launch the Jupyter Notebook below and follow the instructions. ...