MediaPipe Video Masker
This is a project demonstrating how to use Google's MediaPipe to a) mask facial and other personal identiable information (PII) in a video and b) export estimated facial, hand, and body pose features. This work is based on
- Google MediaPipe
- Masked-Piper
- blogs such as https://pysource.com/2021/05/21/blur-faces-in-real-time-with-opencv-mediapipe-and-python/ and https://sefiks.com/2022/01/14/deep-face-detection-with-mediapipe/
Installation and running
Recommended to test this with docker. This demo automatically processes all the videos in the /Input_Videos directory in the docker container and saves the output video and feature data in output directories. It prints out some stats as it processes.
docker build -t mediapipevideomasker .
# run with the default examples
docker run --rm -it mediapipevideomasker
# if you have video files in the "./input" folder on your host machine
# and you want the output to be in "./output"
# Note: for Windows machine only. Replace %cd% with ${PWD} or $(PWD) for Linux
docker run --rm -it -v %cd%/input:/Input_Videos -v %cd%/output:/Output_MaskedVideos -v %cd%/output:/Output_TimeSeries mediapipevideomasker
On my 5-year-old PC the docker runs about 2X real time, depending on the video size and the masking options.
Options
Look into the config dict in mpvideomasker.py. You can control
- whether to mask the face or the whole body
- whether to show pose and facial features
To make changes, you will have to modify the mpvideomasker.py file at the moment, until we implement a proper way to pass options via environment variables or other means.