This project provides a Streamlit-based web application for automatic detection and blurring of sensitive content in videos using a YOLOv8 model. The system is designed for privacy protection, such as anonymizing identity documents, passports, and credit cards.
Features include:
- Automatic Object Detection: Utilizes a YOLOv8 model finetuned on the MIDV-500 dataset for robust detection of sensitive content.
- Multiple Blur Types: Supports various blur/censoring effects, including: Pixelate, Gaussian blur, Motion blur, Blackout, Whiteout, Mosaic
- Object Tracking: Tracks detected objects across frames for consistent blurring, even when objects move.
- Temporal Smoothing: Smooths bounding box positions over time to reduce jitter and improve visual quality.
- Customizable Parameters: Easily adjust blur strength, detection confidence, tracking distance, and smoothing factor via the sidebar.
- Progress Feedback: Real-time progress bar and status updates during video processing.
- Downloadable Results: Download the processed, blurred video directly from the web interface.
Model Training:
- Base Model: YOLOv8n-pt
- Finetuning Dataset: MIDV-500 (identity documents, license plates, numbers)
- Training Epochs: 30
Below are all dependecies used:
- streamlit
- opencv-python
- numpy
- ultralytics
- pyyaml
- scipy
- imageio-ffmpeg
Built With
- python
- ultralytics
Log in or sign up for Devpost to join the conversation.