Skip to main content
Version: 2.0 (alpha)

Choose a tracking model

The tracker finds your body's joints in each video frame. FreeMoCap offers two. Choose one under Mocap Settings > Detector.

RTMPose (default)MediaPipe
AccuracyHigherLower
SpeedSlower, much faster with a GPUFaster, runs well on any processor
Blender exportNot yetYes
OptionsDefault, High Res, or Fast modelHeavy, Full, or Lite model

Use RTMPose for the most accurate data, unless you need Blender.

Use MediaPipe if you want a Blender scene, or if processing is too slow on your computer.

The Detector settings in Mocap Settings with RTMPose selected, showing the Default, High Res and Fast model options and a confidence threshold
With RTMPose selected, choose a Default, High Res or Fast model.FreeMoCap 2.0.0-alpha.25
The Detector settings in Mocap Settings with MediaPipe selected, showing the Heavy, Full and Lite pose model sizes and three confidence settings
With MediaPipe selected, choose a Heavy, Full or Lite pose model.FreeMoCap 2.0.0-alpha.25

Matching the tracker to your task​

From FreeMoCap's validation studies (details in accuracy, validity, and limits):

  • Walking and gait: RTMPose was the most consistent.
  • Balance and standing still: MediaPipe's built-in smoothing made it better at picking up small amounts of sway.
  • Fast movements: MediaPipe's smoothing can make it lag behind quick motion, so prefer RTMPose.

When neither tracker works​

Standard trackers can fail on bodies they weren't trained on, such as a person with a prosthetic limb. A custom-trained model can work in these cases; see image tracking and pose models.

Next steps​