This is the implementation of VGG-11 model using PyTorch. The goal is to build a VGG model from scratch using PyTorch modules which is then trained on CIFAR-10 Dataset. After training for 20 epochs, the accuracy for testing set is around 80%. The parameters are then saved for practicing purposes.
VGG stands for Visual Geometry Group. It is a Convolutional Neural Network(CNN) architecture with convolution blocks followed by fully-connected layers. VGG-11, for example, has 8 convolutional layers and 3 fully-connected layers, hence it is called VGG-11. The original paper can be found here.
CIFAR-10 dataset consists 60000 32x32 RGB images of 10 different classes. There are 50000 training images and 10000 test images. The categories are airplane, automobile, bird, cat, deer, dog, frog, horse, ship, truck. This dataset can be downloaded here.