Closed XJTU-Haolin closed 7 months ago
I have the same question. Did you solve it?
@XJTU-Haolin d_bound
indicates the height range of vehicles compared to a reference ground plane. In the BEVHeight codebase for the Rope3D dataset, the reference ground plane closely matches the real one. For the DAIR-V2X dataset, it's the horizontal plane of the Virtual LiDAR system, positioned higher than the actual ground.
After comparing configs between DAIR and Rope3D, I found a big difference in "d_bound" of the Backbone config. I am not sure what is the exact meaning of "d_bound" (I guess the parameters are related to "height") and why different values are predefined for two datasets. How do we judge and define the values of "d_bound" for a new scene or custom dataset? depends on sensors, or camera pose?
Thanks for your time!