Deepfake Forensics & XAI Engine

An enterprise-grade, multi-modal platform using 15 biological, physical, and acoustic pipelines to detect deepfakes with Explainable AI and generate court-ready PDF reports.

Comment

posted an update —

Major Evolution: Deepfake Forensics & XAI Engine v2.0 is Live!

We’ve just pushed a massive architectural and UI/UX overhaul to the Deepfake Forensics & XAI Engine, evolving it from a standard detection script into a robust, court-ready digital forensics platform.

Here’s a log of our biggest breakthroughs in this release:

1. True Explainable AI: Absolute Probability Impact

Transparency is everything in forensics. Previously, our SHAP (SHapley Additive exPlanations) metrics relied on relative normalization, which could mathematically misrepresent minor artifacts.

We’ve completely refactored the XAI engine to calculate Absolute Probability Impact. Now, the system explicitly tells you exactly how much a specific feature (like "Facial Geometry" or "Error Level Analysis") shifted the final neural network verdict.

2. Court-Ready Forensic PDF Reports

Our backend now synthesizes a breathtaking, multi-page Forensic PDF Report the moment an analysis hits 100%. We engineered a dynamic 2x2 mathematical grid layout to handle over 30 sensor outputs.

To make the reports infinitely scalable and professional, we implemented a dynamic categorical hierarchy. Exhibits are now cleanly grouped under bolded headers:

  • NEURAL ATTENTION & ELA
  • FREQUENCY DOMAIN ANALYSIS
  • FACIAL GEOMETRY & TEXTURE
  • TEMPORAL, AUDIO & MOTION

A peek at our dynamic PDF layout logic:

def gallery_section(title_text):
    nonlocal col_index, max_row_h
    # Auto-align the grid and trigger page breaks seamlessly
    if col_index != 0:
        pdf.set_y(pdf.current_row_y + max_row_h + 15)
        col_index = 0
    if pdf.get_y() > 250:
        pdf.add_page()

    pdf.set_font("Arial", 'B', 11)
    pdf.set_text_color(100, 100, 100)
    pdf.cell(0, 8, title_text.upper(), 'B', 1, 'L')
    pdf.current_row_y = pdf.get_y()

Log in or sign up for Devpost to join the conversation.