ML-Project

This project contains the code for a machine learning model that classifies images as human or AI-generated. It consists of three distinct models, a Convolutional Neural Network (CNN), a Support Vector Machine (SVM), and a Generative Adversarial Network (GAN). It is NOT an ensemble model, but an investigation into the performance of different models on the same dataset.

• main.py : Main script to run the project
• src : Source Folder
  • cnn : CNN model
    • CNNmodel.py : Primary code for the CNN model, containing preprocessing, training, and testing functions
  • svm : SVM model
    • model.py: Primary code for the SVM model, containing preprocessing, training, and testing functions
  • gan : GAN model
    • model_gan.py : Primary code for the GAN model, containing preprocessing, training, and testing functions
    • model : Folder containing the GAN model
    • images : Folder containing the GAN generated images for each epoch
• datasets : Folder containing the dataset utilities for training and testing
  • fake_dataset.py : Class representing the streamed fake dataset - identical data types as real dataset
  • real_dataset.py : Class representing the streamed real dataset - identical data types as fake dataset
  • Train_GCC-training.tsv : Training dataset, consists of URLs and captions
  • Validation_GCC-1.1.0-Validation.tsv : Validation dataset, consists of URLs and captions
  • ataset_InfImagine___fake_image_dataset_default_0.0.0_4edf8865bd54ae2acf4708b709b74b6086b61cba.lock : Fake image dataset, small file containing information retrieved from the HuggingFace datasets library
  • test : Folder containing a 20k test dataset, downloaded and saved from the full dataset
  • train : Folder containing a 100k train dataset, downloaded and saved from the full dataset

Additional files in the root directory include

• requirements.txt : Required packages for the project
• README.md : This file!
• job.sbatch : Slurm job script for running the project on the ICE cluster
• Report-XXXXXXX.out : Output file from running the project on the ICE cluster
• gitattributes : Git LFS file for tracking large files - in our case the training TSV
• gitignore : Generic gitignore

Report documentation is found in the astro-docs branch, with content served on a GH pages site, available here