Convert to physical units

[kjmiller944]

Hi, I was hoping that someone can help me to understand the physical units of the predictions. The paper says that the predictions are in the reference frame of the first camera. What then are the units of depth map--even if they are in a normalized space? A follow-on question is then, can I convert from this normalized space into real-world space by using a camera calibration file that has the actual focal length and pixel pitches of the first camera?

This project is pretty much the coolest thing I've ever seen in computer vision. My jaw dropped seeing this presented at CVPR last year.

[jytime]

Hi we predict the results in a normalized space, where you can find more details in:

#64

If you have the actual focal length and pixel pitches of the first camera, yes you can convert the prediction back to real world!