CNNs

This repo contains various scripts produced during my bachelors graduation internship. These include the scripts for six CNNs to train and test, image preprocessing, and post-training analysis.

The CNNs can be trained to perform binary classification tasks on a set of input images

Table of contents

• Requirements
• CNN testing
• CNN training
• WIP
• Post training analysis
• Preprocessing
• Acknowledgements

Requirements

All models were run and trained on an in-house anaconda environment. This environment mirrors the freely available pytorch2 environment. Additionally, wandb.ai is required for logging the training and validation metrics produced by the CNNs

CNN testing

The CNN scripts within this folder are capable of loading the models produced during training and running them on a test set or functional dataset in order to perform predictions with the trained models.

CNN training

Here, the six CNNs AlexNet, DenseNet121, ConvNeXt Tiny, InceptionV3, ResNet50, and ShuffleNet can be found. All scripts are in working order and capable of performing binary classification tasks. These CNNs require a csv file containing the file path and hyperparameters for the model. An example file can be found within this folder.

Post training analysis

This folder contains all scripts that can be utilised after training or testing of the models has been completed

Preprocessing

In this folder, various scripts can be found that aid in the preprocessing of input data, before training or testing the CNNs.

Acknowledgements

• AlexNet
• DenseNet121
• ConvNeXt Tiny
• InceptionV3
• PyTorch
• ResNet50
• ShuffleNet
• wandb.ai