Tracefy is an advanced AI-powered system that transforms rough criminal sketches and descriptive text prompts into realistic facial images. Built for forensic and law enforcement applications, Tracefy leverages cutting-edge generative models to modernize suspect identification and accelerate investigations.
Tracefy bridges the gap between victim/witness memory and actionable visual evidence using a multi-stage AI pipeline involving sketch refinement and text-guided image generation.
Input: User-drawn outline sketch + natural language description
Output: High-resolution photorealistic facial image
Demo.Video-Tracefy.mp4
-
Sketch-to-Image Translation
Converts rough outlines into high-resolution face images. -
Dual Conditioning
Fuses geometry (sketch) with semantic cues (text) to guide image generation. -
Real-Time Inference
GPU-accelerated pipeline for fast, web-based generation. -
User Interface
React.js frontend with secure, role-based access (RBAC). -
Bias Mitigation
Built-in fairness monitoring to ensure demographic balance. -
Law Enforcement Tools
Secure dashboard to manage sketches, generation history, and retrievals.
-
Sketch Refinement
Uses a CycleGAN-based model for line cleanup and style normalization. -
Image Generation
Integrates refined sketch and user prompt using:- FLUX.1-dev (Latent Diffusion Model)
- ControlNet-Union-Pro-2.0 (structure-aware synthesis)
-
Prompt Iteration
Interactive session flow for refining and improving generated faces.
| Task | Model |
|---|---|
| Sketch Cleanup | CycleGAN |
| Image Generation | FLUX.1-dev + ControlNet |
| Fine-tuning | LoRA (Low-Rank Adaptation) |
| Layer | Technology |
|---|---|
| Frontend | React.js |
| Backend | Flask (Python) |
| ML Framework | PyTorch, TensorFlow |
| Data Storage | MySQL, MongoDB |
| Model Infra | Google Colab, Hugging Face |
- User Authentication (RBAC)
- Sketch Upload & Prompt Input
- Sketch Refinement Preview
- Face Generation & Iteration
- Image History & Download
- Law Enforcement Dashboard
| Metric | Purpose |
|---|---|
| Canny Edge Similarity | Structural accuracy measurement |
| Perceptual Hash (pHash) | Visual resemblance scoring |
| Hamming Distance | Quantifies binary differences in hashes |
Tracefy is a B2G (Business-to-Government) solution. Target sectors include:
- Police departments and forensic teams
- Investigative journalism units
- Security and surveillance technology providers
- Mobile App Deployment
- Acquisition of South Asian Facial Dataset
- Academic & Institutional Collaborations
- Integration with National Criminal Databases (e.g. NADRA)
Tracefy is developed for academic and public safety purposes. For licensing or usage inquiries, please contact the authors.