Regarding interpretation of 3d coordinates of Joints.

[amirkhn33]

I'm interested in doing some inferences on top of 3d co-ordinates generated for each joint.
For instance, I want to calculate:

1. the angle formed between the shoulder, elbow and wrist.
2. translation of a point between two frames.

My observations are:

1. The angles calculated are not even remotely close to their actual value. I'm using simple co-ordinate geometry for calculating angles from 3d points.
2. I observed axes are not on the same scale. I measured the height of a person (distance between the head and centre of feet) and translation of a point (head) between two frames but there is a huge difference in resultant values despite the actual distances being approximately equal.

Am I missing something?

[una-dinosauria]

Hi @amirkhn33,

It is very hard to tell whether you are "missing something", since you don't provide any code nor numbers that you are observing.

That said, I would suggest you use our model trained on ground truth 2d detentions, which has the best performance and is less jittery from frame to frame.

Cheers,

[una-dinosauria]

Closing for lack of activity. Please reopen if you still have issues.