search

dvlab-research / 3D-Box-Segment-Anything

We extend Segment Anything to 3D perception by combining it with VoxelNeXt.
538 stars 25 forks source link

VoxelNext question? #7

Closed TmacTmac1992 closed 1 year ago

TmacTmac1992 commented 1 year ago

Hi, can VoxelNext be replaced with other pretrained 3d detector(i.e. CenterPoint, BEVFusion)?

yukang2017 commented 1 year ago

Hi,

I think it needs fully sparse style network. The dense head network do not have voxel-to-box pair. I think FSD might be a substitute.

Regards, Yukang Chen