MobileNet Training on CIFAR-100

This project demonstrates the training of a MobileNetV2 model on the CIFAR-100 dataset. It includes functionality for creating balanced subsets, defining data loaders, and training with different optimizers.

Features

• Balanced Dataset Subset Creation: Sample subsets with equal representation of each class.
• Data Augmentation and Normalization: Resize, normalize, and prepare the data for training.
• Model Definition: MobileNetV2 with custom classification layers tailored to the CIFAR-100 dataset.
• Training and Evaluation: Train the model with various optimization algorithms and evaluate performance on validation and test datasets.

Usage

1. Dataset Preparation

The notebook automatically downloads and prepares the CIFAR-100 dataset. A balanced subset is created for training and testing.

2. Training

The model is trained on only 10% of the CIFAR-100 dataset to demonstrate the workflow. As a result, the accuracy is expected to be low. For better performance, consider training the model on the full dataset.

Run the training loop using different optimization algorithms. Supported optimizers include:

• SGD
• SGD with Momentum
• RMSprop
• Adadelta
• Adagrad
• Adam

3. Metrics and Evaluation

The notebook tracks:

• Training and validation losses
• Training and validation accuracies
• Final test accuracy

4. Results

Evaluate the performance of the trained model on the test dataset.