| Detailed description |
|
|
I think that this is similar to a prior bug report related to the Inverse Kinematics Tool. OpenSim 4.0 appears to include unselected markers in when calculating the sum of squared errors in the RMS error calculation. However, it does not add those markers to the divisor (number of markers) when taking the mean.
E.g: When running the same scaling operation with to sets of Static Pose Weights, we get the following:
1. 18 markers selected; RMS ~0.175m
2. same 18 + 4 = 22 markers selected (additional markers are not fixed and have zero weight); RMS ~ 0.159m.
The ratio of RMS errors (~0.90) is roughly equal to the ratio of the square root of marker counts (sqrt(18/22)).
Running an identical scaling operation - same setup file, markerset, measurement set, TRC files, generic OSIM model - in OpenSim 3.3, RMS errors are approximately 0.02m.
Notes:
- All non-fixed markers in (1) were unselected.
- Max error also includes unselected markers.
- IK returned RMS error less than 0.02m, despite the high scaling errors.
Happy to provide images upon request.
Thanks!
Best,
Michael |
|
