deep-learning-project's Introduction

Deep-Learning-Project

This project implements U-Net and DeepLabV3_MobileNetV3 deep learning models for image segmentation specific to the Diabetic Foot Ulcer dataset provided on https://dfuc2022.grand-challenge.org.

These models have been built using Python version 3.9.16 and PyTorch version 1.12.1.

Directory Guidance

Assignment.xlsx shows the results of the model training tasks based on experiments.

Notebooks:

main.ipynb shows the full building and running to produce best model.
augmentation.ipynb gives an overview of the different image augmentation methods used in this project.
exploration.ipynb explores the dataset and gives an overview of statistics.
modelTesting.ipynb evaluates the baseline and best models that have been trialled.
train_test_split.ipynb shows an example of how the images can be split into training and testing lists.
csv files save of the descriptive statistics from the exploration.ipynb

Code - OOP Programming of the Project:

main.py running this file will train the model and produce results saved in "./models"
loss.py defines the dice loss function, the IOU evaluation metric and loads BCE with Logits Loss from Pytorch.
dataset.py the torch dataset for loading and normalising images.
model.py builds a U-Net convolutional neural network and load DeepLabV3_MobileNetV3 from PyTorch.
optimiser.py loads the SGD and ADAM optimisers from PyTorch.
readFiles.py reads the directories of the images saved in "./dfuc2022/"
training.py provides the training loop for training and evaluation of validation data.