Closed KengChiLiu closed 6 years ago
We didn't compute but you could use the camera poses also provided in the dataset to generate height from the ground plane.
Sure~ @ankurhanda By the way, just for curiosity, why did you add H and A for your work? Do you think computing H and A is practical in real world ?
I think it is sensible to use only if you can start your reconstruction looknig at the floor first and are able to track the camera pose wrt that. However, this will drift over time and you will need to do some pose graph optimisation with a constraint that the points on the floor must have z = 0. See SLAM++ it uses that constraint. Otherwise, working with depth is pretty much the best you can do without going for reconstruction.
Hope this helps.
@ankurhanda I see. Pretty much the same as I thought. Thank you so much~
Dear @jmccormac , @ankurhanda ,
How can I get the DHA feature for SceneNetRGBD ?
Thank you~