Overview

• The CIFAR-10 dataset has 10 categories: airplane, automobile, bird, cat, deer, dog, frog, horse, ship, and truck.
• Used Google Colab, a cloud-based platform, to train Convolutional Neural Network (CNN) models with and without transfer learning.
• Transfer learning CNN model was trained on top of the VGG16 model of Keras.
• This project was a 10-class classification problem.

Methodology

• Polished the dataset by loading the data set from CIFAR-10 and splitting it into train and test data.
• Upsampled the images from 32x32 pixels to 64x64 pixels to improve the performance of the model.
• CNN model without transfer learning, applied four Convolutional layers with 32, 32, 64 and 64 neurons (kernels) respectively.
• Applied kernel size of 3x3 and had two Max Pooling layers with pool sizes of 2x2 and 1x1 stride to reduce variances and computations for our model.
• Had a Dense layer with 512 neurons and a Dropout layer with a probability of 50% to drop a neuron.

Result

Wihtout transfer model:

With trasnfer model: