Skip to content

AI-powered media forensics dashbaord that uses VQA, NLI, and ELA to detect deepfakes, tampering, and logical contradictions between an image and its caption.

Notifications You must be signed in to change notification settings

ZohaibHassan16/ClaimTrace

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 

History

30 Commits
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

ClaimTrace: Multimodal AI Forensics

ClaimTrace is a media forensics platform designed to combat misinformation and deepfakes. It utilizes a multimodal AI pipeline to cross-reference visual evidence against textual claims, detecting semantic inconsistencies, AI-generated content and digital tampering.

Capabilities

  • Synthetic Media Detection: Implements Vision Transformer (ViT) to identify AI-generated imagery.

  • Semantic Consistency Verification: leverages BLIP-2 (Vision-Language Model) to perform Zero-Shot Visual Question Answering (VQA). It "sees" the image context and mathematically compares it against the text caption using Natural Language Interface (NLI) cross-encoder.

  • Digital Forensics (ELA): Performs Error Level Analysis to detect compression anomalies, highlighting regions where pixels have been manually altered or spliced.

  • Hybrid Inference Engine: Features a device-agnostic architecture that automatically optimizes for 4-bit Quantization on NVIDIA GPUs or bfloat16 precision on CPUs, ensuring deployment flexibility across edge devices and cloud servers.

System Architecture

The system follows a microservices architecture, containerized for scalibility.

ClaimTrace Architecture Flowchart

Stack

  • Core: Python 3.11, PyTorch

  • Models: Salesforce/blip2-opt-2.7b (Vision-Language), cross-encoder/nli-distilroberta-base (Logic), umm-maybe/AI-image-detector (Synthetic Detection).

  • API Framework: FastAPI

  • Deployment: Docker

  • Optimization: bitsandbytes (4-bit Quantization), accelerate (Device mapping).

Quick Start (Docker)

This application is fully containerized to resolve dependency conflicts.

1. Build the Container

docker build -t claimtrace .

2. Run the Forensic Engine

docker run -p 8000:8000 claimtrace

Access the dashboard at http://localhost:8000 and the API documentation at http://localhost:8000/docs.

Manual Installation (Local Dev)

For development without Docker:

git clone https://github.com/ZohaibHassan16/ClaimTrace.git
cd ClaimTrace

python -m venv venv
source venv/bin/activate # Windows: .\venv\Scripts\activate


pip install -r requirements.txt # Use Python -m if on windows


python -m uvicorn app.main:app --reload

About

AI-powered media forensics dashbaord that uses VQA, NLI, and ELA to detect deepfakes, tampering, and logical contradictions between an image and its caption.

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published