I recently encountered significant discrepancies in the calculated Euler angles (pitch, yaw, and roll) when using both Mediapipe and Openface for the same set of images. Despite using the provided code from their respective websites, I observed notable differences in the resulting angle values, sometimes varying by a considerable margin.

My question pertains to understanding the underlying mechanisms employed by these tools in calculating rotation angles. While I anticipated some variance due to differences in algorithms or methodologies, the magnitude of the differences observed seems unexpected.

Any insights or suggestions on how to reconcile these differences would be greatly appreciated.
Thank you for your time and expertise.

Best,
AB