Image classifiers on the CIFAR 10 subset

About the project

The purpose of the project is to implement convolutional neural network (CNN) and multilayer perceptron (MLP) image classifiers and compare the performance of them on a subset of the CIFAR 10 dataset.

Datasets

250/200 split subset of CIFAR-10 is loaded from keras.datasets for train/test.

Implemented models

1. CNN model on raw images

2. MLP model on raw images

3. MLP model on extracted features from pre-trained VGG19

   It extracts image features from VGG19 layer block5_conv4 and uses the representative features.

4. MLP model on extracted features from pre-trained VGG19 after resizing

   It upscales the original raw images (32x32) into 224x224 before feature extraction.

These models are trained with the same training settings.