Live Demo

MaskAnyone in Action

See how SYNAPSIS transforms identifiable audiovisual data into privacy-preserving research material while maintaining analytical value.

MaskAnyone

Several people, one room, each masked

A clinical training simulation. MaskAnyone follows each person separately with SAM2 and masks them while keeping their pose and hand tracking, even when people cross and overlap. (The patient is a training manikin.)

Masked
Demo from the MaskAnyone repository.
Same clip, four strategies

What masking looks like

One public TED talk, masked four ways. The coloured skeleton is pose tracking run on the masked video: it shows how much movement data survives each strategy. Clips from the MaskBench benchmark.

Blur
Gaussian blur over the whole person. Posture and gesture stay readable.
Pixelation
Block-based obscuring. Familiar and fast; how much it hides depends on the block size.
Contours
Only the outline is kept. Appearance goes; the body shape and movement remain.
Solid fill
A flat silhouette removes appearance entirely, yet pose is still recoverable.
Masking lite

Masked-Piper: hide the person, keep the signal

Masked-Piper masks face and body but keeps the face mesh, hand and body tracking drawn over the mask, so gesture, gaze direction and facial movement stay analysable. It runs on an ordinary laptop.

One run, twelve outputs
The same clip rendered in every output mode: skeleton only, face or body masked, face or body blurred, finger traces, and more. From the EnvisionBox Masked-Piper module (MIT licence).
Gesture
Hands tracked finger by finger while the person is fully masked.
Face mesh
The face mesh keeps expression and mouth movement without showing the face.
Full body
Whole-body masking with the pose skeleton kept, on a public TED talk.

Clips from Wim Pouw’s TowardsMultimodalOpenScience examples, accompanying Owoyele et al. (2022), SoftwareX.

Try it yourself

Open tools, free to use

Masking lite: Masked-Piper

A step-by-step EnvisionBox module. Runs on any laptop, keeps body and face kinematics.

EnvisionBox module

MaskAnyone

Precision masking in a browser interface, run locally or on a department server.

Code & install guide

MaskBench

All benchmark videos and results: how pose estimation holds up under each masking strategy.

Preprint & results
Methods

Available Masking Techniques

Pixelation

Block-based face obscuring. Fast, simple, widely understood.

Privacy: High

Blur

Gaussian blur over facial regions. Adjustable intensity.

Privacy: Medium-High

Face Masking

Mask the face only, keeping body movement and gesture visible. See examples on EnvisionBox.

Privacy: Medium-High | Utility: High

Skeleton Only

Extract pose data, render skeleton visualization only.

Privacy: Maximum
Data Extraction

Pose & Kinematic Data

Beyond masking, SYNAPSIS extracts skeletal pose data from videos - enabling gesture analysis, movement studies, and behavioral research without exposing identity.

  • 33 body keypoints (MediaPipe) or 25 (OpenPose)
  • Hand landmarks (21 points per hand)
  • Export to JSON, CSV for analysis in R/Python
  • Blendshape extraction for facial expression analysis 
{
  "frame": 42,
  "timestamp": 1.4,
  "pose": {
    "nose": [0.52, 0.31, 0.98],
    "left_shoulder": [0.61, 0.48, 0.95],
    "right_shoulder": [0.43, 0.47, 0.96],
    "left_wrist": [0.71, 0.62, 0.89],
    "right_wrist": [0.33, 0.58, 0.91]
    // ... 33 keypoints total
  },
  "hands": {
    "left": [...],
    "right": [...]
  }
}
Sample pose data output (JSON)
Live Platform

MaskAnyone Demo

MaskAnyone is deployed on Radboud University infrastructure. Access is currently restricted to project partners and pilot participants.

Request Access View Training Install it yourself

Hosted on Radboud Ponyland cluster. Contact us to arrange a guided demo session.

Want to Learn More?

Explore our training materials or get in touch to discuss deploying MaskAnyone at your institution.

Partner Network Resources