Hi there!
I am working with a camera mounted to a UR10e gripper. The application requires very high precision, so I am performing a handEye calibration to measure the relatives angles between the TCP and the camera. They are suppose to be [0, 0, 180], but because the mounting is not precise, I know that the error is around 1° in each coordinate. 
To do the calibration I need to get the UR rotation matrix from the axis-angle notation. I created a Polyscope program to extract the robot poses. It gets them in axis-angle notation, in radians. However, If you check the axis-angle notation on the Polyscope (under move) and you compare with the script values of that same point, you will see a difference. From another post, I understood that the Polyscope values are scaled following the next formula: 

p[x,y,z] - script values
scale = 1 - 2*PI/length(p[x,y,x])
polyscope_values = [scale*x, scale*y, scale*z]

I took 5 different poses on the UR and realized, that this formula is correct. 
But, which one should I use to get the rotation matrix?

Thank you for the suggestion

For our specific application is not possible to use a Wrist camera. We are working with a RealSense D415 stereo camera to get depth information of the scenes. We are working in a high unstructured environment, so I don´t think that we could change that.

The part of the handEye calibration is fixed already. I am anyway curious about the difference in the axis-angle values and the scaled issue. 

Kind regards,