Depth recovery issue

[aim-uofa]

Depth recovery issue

[guangkaixu]

@jing98tian Hi, if you can obtain some sparse depth points, you may be interested in this work: https://github.com/guangkaixu/BoostingDepth, which leverages sparse points to ensure depth consistency between frames.

[YvanYin]

Hi, I have not tested the scale consistency problem. Did you use our method on a video?

[jing98tian]

Yes, it's tested on a video in realtime. And i use the relsense d435i to get the true depth value to reacover the scale parameter.
In the case of camera motion, the scale gray parameter is not robust enough. Should the true depth value be used to estimate the scale recovery parameter all the time?

[YvanYin]

I believe that you can use some sparse points that have known depth to recover the scale and shift for the predictions for each frame. Our method doesn't optimize the scale consistency.