🔍 Signal Pro // Forensic Engine v3.0

Advanced Forensic Analysis Console & Deepfake Detection Platform

Uncover the Truth in Media.
Upload suspect videos → Get real-time AI-driven inference, frame-by-frame visualization, and detailed PDF reporting using temporal and spatial analysis.

---

📸 Screenshots

Dashboard	Real-Time Inference	PDF Reporting
System monitoring + Case history	Live bounding boxes + Metrics	Automated, Institutional-grade reports

---

✨ Features

🎯 Core Intelligence

• Multi-Modal Analysis — Unifies spatial artifacts, frequency noise signatures, and temporal consistencies to evaluate the media.
• Explainable Metrics — Clear percentage displays mapping directly to "REAL", "SUSPICIOUS", or "FAKE" verdicts.
• Deepfake Simulation — Integrates Xception (spatial), SRM (noise), and BiLSTM (temporal) for robust detection capabilities.

🔬 Deep Analysis

• Temporal Inconsistency Detection — Uses BiLSTM networks to track unnatural temporal changes across frames over time.
• Noise Signature Extraction — Steganalysis Rich Model (SRM) filter captures invisible manipulations and generative noise.
• Face Bounding Box Visualization — Real-time Red/Green bounding boxes indicate localized anomalies per frame.

🚀 Forensic Workflow

• Automated Reporting Engine — Automatically generates professional PDF reports with executive summaries and visual evidence breakdowns using ReportLab.
• Case History Tracking — Analysis history is persistently saved to a local JSON database and reviewable in real time.
• System Monitoring — Live monitoring of model load statuses, GPU hardware capability (CUDA), and software environment metrics.

🖥️ Premium Dashboard

• Modern PyQt5 Interface — Sleek, dark-themed interface built on purely native widgets, eliminating heavy dependencies.
• Dynamic Animations — Smooth progress bars and fluid transitions across panels resulting in a premium native feel.
• Executable Deployment — Configured for one-click compilation as a standalone Windows .exe via PyInstaller.

---

🏗️ Architecture

┌─────────────────────────────────────────────────────────┐
│                    PyQt5 Frontend                       │
│              (ui/main_window.py — GUI)                  │
└─────────────────────┬───────────────────────────────────┘
                      │ Internal Multi-Threading
┌─────────────────────▼───────────────────────────────────┐
│                   Inference Worker                      │
│                (ui/workers.py — Logic)                  │
├─────────────────────────────────────────────────────────┤
│                Engine Coordination Layer                │
│             Frame Extraction & Inference                │
├─────────────────────────────────────────────────────────┤
│                    Model Pipeline                       │
│  ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│  │ Face Detect  │ │   Cropping   │ │ Preprocessing    │ │
│  └──────────────┘ └──────────────┘ └──────────────────┘ │
│  ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│  │ Spatial (CNN)│ │ Freq (SRM)   │ │ Temporal (LSTM)  │ │
│  └──────────────┘ └──────────────┘ └──────────────────┘ │
│  ┌──────────────┐ ┌──────────────┐ ┌──────────────────┐ │
│  │  Score Fusion│ │  Reporting   │ │ Persistence      │ │
│  └──────────────┘ └──────────────┘ └──────────────────┘ │
├─────────────────────────────────────────────────────────┤
│                     Core Layer                          │
│  ┌──────────────────┐  ┌─────────────────────────────┐  │
│  │ Model Wrapper    │  │ Export Service (PDF)        │  │
│  │ (models.py)      │  │ (services/export_service.py)│  │
│  └──────────────────┘  └─────────────────────────────┘  │
└─────────────────────────────────────────────────────────┘

---

🛠️ Tech Stack

Layer	Technology
Frontend UI	PyQt5, SVG Material Icons
Logic & Concurrency	Python Custom Threads (QThread)
Computer Vision	OpenCV (cv2), facenet-pytorch MTCNN
Deep Learning Base	PyTorch, torchvision
Models	Xception (Spatial), Pre-filtered Xception (SRM), BiLSTM (Temporal)
Data Mgt & Reporing	JSON, ReportLab (PDF Parsing)
Deployment	PyInstaller (Standalone Windows exe)

---

📁 Project Structure

forensic_ai_desktop/
├── app.py                    # Application Entry Point
├── build_exe.py              # PyInstaller Build Script
├── analysis_history.json     # Persisted local database map
├── requirements.txt          # Deep Learning & GUI Packages
│
├── core/
│   ├── config.py             # Hyperparameters, Paths, & Globals
│   └── theme.py              # (Legacy) Color & Padding Specs
│
├── ui/
│   ├── main_window.py        # Master Dashboard & Event loops
│   ├── workers.py            # Async Model Load & Video Inference
│   ├── widgets.py            # Modern TitleBar, ProgressBars
│   ├── styles.py             # Application Qt Stylesheet
│   └── icons.py              # Material Design SVG Assets
│
├── services/
│   └── export_service.py     # PDF Report Engine via ReportLab
│
├── models.py                 # PyTorch Model Subclasses
├── *.pth                     # Pretrained Weights (LFS Tracked)
├── reports/                  # Generated PDF Output Directory
└── processed_faces/          # Debug frame crops

---

🚀 Getting Started

Prerequisites

• Python 3.8+
• Git with Git LFS installed (Crucial for *.pth models)
• NVIDIA GPU with CUDA support strongly recommended.

Installation

# 1. Clone the repository
git clone https://github.com/<your-username>/forensic-ai-desktop.git
cd forensic-ai-desktop

# 2. Pull large LFS Objects (Wait for download completion)
git lfs install
git lfs pull

# 3. Create a virtual environment
python -m venv .venv

# 4. Activate the virtual environment
# Windows:
.venv\Scripts\activate
# macOS/Linux:
source .venv/bin/activate

# 5. Install dependencies
pip install -r requirements.txt

Running the Application

Option 1: Native Python

python app.py

Option 2: Build as an Executable (Windows)

python build_exe.py

This drops the standalone Forensic_AI_Console.exe in the dist folder.

---

📖 How It Works

Video Upload (.mp4/.avi)
    │
    ▼
Frame Extraction ──► MTCNN Face Detection ──► Face Cropping
    │
    ▼
CNN Spatial Feature Extraction (Xception)
    │
    ├──► SRM Filter Analysis (Identify high-frequency noise)
    └──► Temporal Assessment (Run feature vectors thru BiLSTM)
            │
            ▼
        Score Fusion Engine
    (Weight Spatial, Freq, Temp)
            │
            ▼
    Final Verdict & Confidence
            │
            ├──► UI Panel Update
            ├──► History Persistence (.json)
            └──► PDF Report Generation

---

🤖 Models & Configurations

Component	Architecture	Weights / Dimensionality
Spatial Engine	Xception	~80 MB
Frequency Engine	SRM + Xception	~80 MB
Temporal Engine	BiLSTM	~15 MB

Note: Models are tracked with Git LFS. If models are absent on run, double-check that git lfs pull fetched the fully sized .pth files.

---

⚙️ Configuration

Key settings modified in core/config.py:

DEVICE          = "cuda:0" if torch.cuda.is_available() else "cpu"
APP_VERSION     = "3.0.0"
IMG_SIZE        = 299      # Expected tensor img dimensions
SEQ_LENGTH      = 10       # Frames required for BiLSTM inference

---

🤝 Contributing

1. Fork the repository
2. Create your feature branch (git checkout -b feature/forensic-engine)
3. Commit your changes (git commit -m 'Enhance Temporal Analysis')
4. Push to the branch (git push origin feature/forensic-engine)
5. Open a Pull Request

---

📄 License

This project is licensed under the MIT License — see the LICENSE file for details.

---

Built with ❤️ using Python, PyQt5 & PyTorch
Signal Pro — Uncover the Truth in Media.