We wanted to share the results of the consistency testing we have done.
The standard deviation for a 1000 image data set was ±1.31 (Left Arm) and ± 1.78 (Right Arm)
How we measured the consistency:
We took the same image and randomly applied the following transformations:
- Image rotation by -5 to 5 degrees
- Random brightness changes
- Random contrast changes
Using these transformations, we had a data set of 1000 images.
For each modified image we applied our AI model and computed the shoulder angles.
For each arm, we produced 1000 angles for each arm.
We calculated their mean and standard deviation. See charts below.