GenImageClass

This project aim is to train classifier in order to distinguish if an image is real or ai generated. We use an Image -> Caption -> Image proccess and then we use this dataset in order to train various classifiers to distinguish the fake ones!

TODO List:

• Investigate more suitable models
• Add Vscode for standarization (formaters)?

Preprocess.py

This script iterates over a dataset an image dataset and standarizes every image to be 1024*1024. It achieves this by either upscaling an image or cropping it.

Florence.py

The florence script is used to parse Actual Images ang generate captions for each one in order to be used later on the image creation process.

Stable Diffusion.py

These scripts use Stable Diffusion 3.5, Stable Diffusion 1.5 and Sana 1.5 in order to generate "fake" images from the captions similar to the actual ones.

Examples

Folder Structure

asset
Codebase
Models
Dataset
├── Captions
│   ├── Train
|   └── Valid
├── Images
|   |
│   ├── Train
|   |   ├── Actual
|   |   ├── sd35
|   |   ├── sd15
|   |   └── sana15
│   └── Valid
|       ├── Actual
|       ├── sd35
|       ├── sd15
|       └── sana15
|
└── Original Dataset
    ├── DIV2K_train_HR
    │   └── DIV2K_train_HR
    └── DIV2K_valid_HR
        └── DIV2K_valid_HR