Objective:

The objective of this assignment is to familiarize you with the concept of transfer learning using

PyTorch. You will modify the classification head of a pre$-$trained MobileNetV$2$ model and train it

on the CIFAR$-100$ dataset. The rest of the model will remain frozen to leverage the pre$-$trained

ImageNet weights. This assignment will help you understand how to harness the power of

pre$-$trained networks for tasks with different datasets and classes.

Dataset:

CIFAR$-100$ consists of $600003232$ color images in $100$ classes, with $600$ images per class.

There are $500$ training images and $100$ testing images per class.

Tasks:

Setup Environment $(10$ marks$)$

Import necessary libraries and load the CIFAR$-100$ dataset using torchvision.

Normalize the dataset with mean and standard deviation: mean $,$

$0\mathrm{.}406,$ std $=[0\mathrm{.}229,0\mathrm{.}224,0\mathrm{.}225].$

You are free to use any other transforms for data pre$-$processing.

Prepare MobileNetV$2(20$ marks$)$

Load a pre$-$trained MobileNetV$2$ model from torchvision.models.

Freeze all layers of the model except the classification head.

Replace the classification head with new layers of your choice to accommodate

$100$ output classes. Justify your design choice.

Training Setup $(20$ marks$)$

Define the loss function and optimizer. You are only allowed to train the

parameters of the new classification head.

Configure the training parameters: batch size, number of epochs, learning rate

$($justify your choices$).$

Training Process $(20$ marks$)$

Implement training and validation loops. Ensure to only update the weights of the

new classification head.

Save checkpoints after every epoch.

Use appropriate device handling $($cpu or cuda$).$

Evaluation and Metrics $(20$ marks$)$

Evaluate the model on the test set.

Calculate and report the following metrics:

Accuracy

Precision, Recall, and F$1-$Score for each class $($report the average over

all classes also.$)$

$$ Provide a confusion matrix of the model predictions.

$6.$ Report $(10$ marks$)$

$$ Provide a detailed report including:

$$ Introduction to the problem and the model architecture.

$$ Explanation of your design choices for the new classification head.

$$ Discussion on the training process, including any challenges faced and

how you overcame them.

$$ Analysis of the results with insights and possible improvements.

$$ Your colab file link.