Redacted

Redacted protects privacy by automatically removing sensitive data from images — from private texts to company documents to passports — so what’s yours stays yours, always.

Comment

Redacted: Privacy-Aware AI Image Detection-Redaction App by CodeTokers

Inspiration

With the rise of social media, users often unknowingly share sensitive information, like license plates or recognizable landmarks, that can disclose their location. Company documents are also often unknowingly leaked in the background of photos. We wanted to create a tool that helps users post safely by detecting and censoring these details automatically.

What it does

  • Redacts texts on documents and slides in the image.
  • Detects license plates in images and censors them to protect privacy.
  • Checks for the presence of recognizable landmarks in photos to alert users about potential location disclosure.
  • Runs on-device in a Flutter app using a YOLO-trained TFLite model.
  • Provides real-time feedback to users before they share images.

How we built it

  • Model Training: Python notebooks (notebooks/) using YOLO for license plate detection.
  • Model Export: YOLO models exported to TFLite format (yolo_export/) for Flutter integration.
  • App: Flutter source code (lib/) for loading the TFLite model, processing images, and providing UI alerts.
  • Censoring & Analysis: Text documents and detected license plates are automatically blurred, and landmark detection is used to warn users.

Challenges we ran into

  • Exporting YOLOv11 models to TFLite while preserving custom layers.
  • Ensuring on-device inference is fast enough for real-time use.
  • Detecting landmarks reliably in diverse environments.
  • Balancing accuracy with mobile resource constraints.

Accomplishments that we're proud of

  • Seamless integration of Python-trained YOLO models into Flutter via TFLite.
  • Real-time license plate censoring and location risk detection.
  • A complete end-to-end workflow from model training to mobile deployment.

What we learned

  • How to train and export YOLO models for mobile use.
  • Techniques for privacy-preserving image analysis.
  • Flutter integration with TFLite and real-time image processing.

What's next for CodeTokers

  • Expand detection to other sensitive content (faces, IDs).
  • Expand text censors to all languages.
  • Improve landmark detection using advanced models for better location alerts.
  • Optimize TFLite models further for faster inference and lower memory usage.
  • Add user customization for privacy settings and alert thresholds.

Built With

Share this project:

Updates