ACM Grand Challenge on Cheapfakes Detection (2023)

This project was developed as part of the ACM Grand Challenge on Cheapfakes Detection (2023). It addresses the critical task of identifying cheapfakes—instances of miscontextualized or misleading media content.

About the Project

This repository contains a Flask application designed to detect fake news, with a particular focus on cheapfakes. The system leverages [brief description of any key technologies or methodologies, e.g., machine learning, natural language processing, etc.] to identify and classify misleading content effectively.

Dataset(COSMOS):

• Train: 160K images, 360K captions
• Validation: 40K images, 90K captions
• Test: 1700 images, 3400 captions (with context annotations)

Key Features

• Binary classification: OOC (label 1) or Not OOC (NOOC, label 0).
• Input: Image with two captions.
• Multi-modal analysis combining image and text.
• Ensemble model technique for classification.
• Context-aware image representation learning.

Getting Started

1. Clone the repo: git clone https://github.com/simR122/final_rep.git
2. Install dependencies: pip install -r requirements.txt
3. Download the dataset using the provided script
4. Visualize data: python dataset_visualizer/visualize.py
5. Train and test for image-text matching:
   • Set up Detectron2 following official instructions
   • Configure utils/config.py
   • Train: python trainer_script.py -m train
   • Evaluate: python trainer_script.py -m eval
6. Test for out-of-context detection: python evaluate_ooc.py

For detailed instructions, refer to the (https://detecting-cheapfakes.github.io/icmr-2024.html).