what is the shape of the virtual points dimension?

[mihudaner]

hi~

I wonder if using CenterPoint changes the input part of the model? Does the virtual point have dimensions other than (x,y,z,r)? If I want to use the kitti data set, and the kitti data set has input (x,y,z,r), does that mean that I need to change the input part of the detection model

[tianweiy]

it does. in addition to xyz, it also gets semantic information about this class (class and instance score in nuscenes case). Also it doesn't perform too well on KITTI, I suggest you checking out related approaches (see #29 (comment))

[mihudaner]

I saw a performance comparison of the pointpillar in the Kitti dataset in the paper. Is this code publicly available? If not, I would like to know what framework it is based on or how to reproduce it. I also want to try it out, but I don't have a clue yet. I hope to reproduce some of it, please (●'◡'●)

[tianweiy]

see #1 and #20

[mihudaner]

thanks(感谢大佬)(●'◡'●)

[mihudaner]

I installed MaskFormer according to the documentation, but why did I keep reporting this error? What folder is this. data? I couldn't find it

[tianweiy]

oh that folder was previously ignored in git. .i just updated the repo and it should work now