How bohdan213/thesis_synthetic_image_detection works
This repository is a research implementation for detecting AI-generated images in misinformation contexts. The principal workflow is to prepare or obtain image datasets, configure and run a classifier notebook for training or inference, and inspect its results. The documented alternatives are a foundation-backbone/MLP classifier and a two-branch foundation-plus-LBP/ResNet-18 classifier. Dataset-generation notebooks are named for several image-generation models. Notebook internals were not sampled, so links reflect README-described architecture and workflow rather than unverified cell-level calls.
Main entry points: foundation-lbp-classifier.ipynb Foundation + LBP classifier · foundation-classifier.ipynb Foundation classifier · generate_flux1_dev.ipynb Flux generation · generate_pixart.ipynb PixArt generation
Components 17
Data preparation
- GenImage data
- Misinfo data
- Flux generation
dataset/dataset-creation-scripts/generate_flux1_dev.ipynb - PixArt generation
dataset/dataset-creation-scripts/generate_pixart.ipynb - SDXL generation
dataset/dataset-creation-scripts/generate_sdxl1.ipynb - Imagen generation
dataset/dataset-creation-scripts/genererate_imagen.ipynb
Detection models
- Foundation classifier
foundation-classifier.ipynb - Foundation + LBP classifier
foundation-lbp-classifier.ipynb - Foundation backbone
foundation-classifier.ipynb - MLP head
foundation-classifier.ipynb - Foundation branch
foundation-lbp-classifier.ipynb - LBP representation
foundation-lbp-classifier.ipynb - ResNet-18 branch
foundation-lbp-classifier.ipynb - Common MLP head
foundation-lbp-classifier.ipynb
Evaluation
- Training and inference results
foundation-classifier.ipynb - Training and inference results
foundation-lbp-classifier.ipynb
Other components
- Researcher
Connections 24
- Researcher → Flux generation: runs
- Researcher → PixArt generation: runs
- Researcher → SDXL generation: runs
- Researcher → Imagen generation: runs
- Flux generation → GenImage data: generates images
- PixArt generation → GenImage data: generates images
- SDXL generation → GenImage data: generates images
- Imagen generation → GenImage data: generates images
- GenImage data → Foundation classifier: training and evaluation data
- GenImage data → Foundation + LBP classifier: training and evaluation data
- Misinfo data → Foundation classifier: evaluation data
- Misinfo data → Foundation + LBP classifier: evaluation data
- Researcher → Foundation classifier: configures and runs
- Researcher → Foundation + LBP classifier: configures and runs
- Foundation classifier → Foundation backbone: uses backbone
- Foundation classifier → MLP head: uses head
- Foundation + LBP classifier → Foundation branch: uses branch
- Foundation + LBP classifier → LBP representation: uses representation
- Foundation + LBP classifier → ResNet-18 branch: uses branch
- Foundation + LBP classifier → Common MLP head: uses head
- Foundation classifier → Training and inference results: produces
- Foundation + LBP classifier → Training and inference results: produces
- Researcher → Training and inference results: reviews
- Researcher → Training and inference results: reviews