Camera calibration requirements for deployment?

Hi!
Thanks for the great work, here we are evaluating Molmoact2 in our setup, facing a problem with calibration:
Wondering if RealSense camera calibration is needed for YAM, the paper doesn't document any intrinsic calibration step.

As MolmoAct2-Think uses Depth Anything V2 for monocular depth, it seems like intrinsics don't matter.
- Is predicted depth metrics or relative?
- Will it be helpful to calibrate the prediction depth with camera depth?

Would be helpful to know camera config protocol, and better to have a recommended setup guide for hardware reproduction.