Real-Time Face and Text Deidentification ¶

Authors: Wendell Hom (NVIDIA), Jonathan McLeod (NVIDIA)
Supported platforms: x86_64, aarch64
Last modified: March 18, 2025
Language: Python
Latest version: 1.0
Minimum Holoscan SDK version: 2.1.0
Tested Holoscan SDK versions: 2.1.0
Contribution metric: Level 2 - Trusted

This sample application demonstrates the use of face and text detection models to do real-time video deidentification. Regions identified to be face or text are blurred out from the final image.

NOTE: This application is a demonstration of real-time face and text deidentification and is not meant to be used in critical applications that has zero error tolerance. The models used in this sample application have limitations, e.g., in detecting faces and text that are partially occluded, in low lighting situations, when there is motion blur, etc.

Models¶

This application uses TAO PeopleNet model from NGC for detecting faces. The model is downloaded when building the application.

For text detection, this application uses EasyOCR python library which uses Character Region Awareness for Text Detection (CRAFT).

Data¶

This application downloads a pre-recorded video from Pexels when the application is built for use with this application. Please review the license terms from Pexels.

NOTE: The user is responsible for checking if the dataset license is fit for the intended purpose.

Input¶

This app currently supports three different input options:

v4l2 compatible input device (default, see V4L2 Support below)
pre-recorded video (see Video Replayer Support below)

Run Instructions¶

V4L2 Support¶

This application supports v4l2 compatible devices as input. To run this application with your v4l2 compatible device, please plug in your input device and run:

./dev_container build_and_run video_deidentification

By default, this application expects the input device to be mounted at /dev/video0. If this is not the case, please update applications/video_deidentification/video_deidentification.yaml and set it to use the corresponding input device before running the application. You can also override the default input device on the command line by running:

./dev_container build_and_run video_deidentification --run_args "--video_device /dev/video0"

Video Replayer Support¶

If you don't have a v4l2 compatible device plugged in, you may also run this application on a pre-recorded video. To launch the application using the Video Stream Replayer as the input source, run:

./dev_container build_and_run video_deidentification --run_args "--source replayer"

Known Issues¶

There is a known issue running this application on IGX w/ iGPU and on Jetson AGX (see #500). The workaround is to update the device to avoid picking up the libnvv4l2.so library.

cd /usr/lib/aarch64-linux-gnu/
ls -l libv4l2.so.0.0.999999
sudo rm libv4l2.so.0.0.999999
sudo ln -s libv4l2.so.0.0.0.0  libv4l2.so.0.0.999999

Real-Time Face and Text Deidentification¶