OVERVIEW This project aims to implement VGGNet, a deep convolutional neural network (CNN) architecture, using the PyTorch framework from scratch. The objective is to efficiently use the PyTorch framework and create a model without using any pre-trained models.
STRUCTURE The project contains the following files:
model.ipynb README.md - A detailed description of the project, how to run the code, dependencies, and reference materials.
DEPENDENCIES This project requires the following dependencies to be installed:
Python 3.6 or later PyTorch 1.8.0 or later Torchvision 0.9.0 or later NumPy 1.19.5 or later
DATASET The dataset used for this project is taken from kaggle. its a dataset with 8 classes each representing an individual avenger. it contains a training and testing set of data. However due to the huge time it took me to train the model I had to cut the dataset to contain only 2 classes. the details of the number of images for each set can be found in the code.
INFERENCES
since the model is a very deep neural network with 16 layers it has a lot of parameters to tune. so it took a very long time to trian just one epoch(45mins) approx.
finally i was able to reduce the dataset size and it took 3.5 hrs to finish 10 epochs. 
in the epochs the train accuracy and test accuracy both are less which is indicative of underfitting
to prevent underfitting its necassary to add more data so that a very complex model like vgg16 can be trained properly.
i also added batchnormaliation and reduced the dropout layers to normalize the input to every layers.
trivial augment wide function was also incorporated to increase the randomness of the training set and to introduce more vareity.