SamsungLabs / imvoxelnet

[WACV2022] ImVoxelNet: Image to Voxels Projection for Monocular and Multi-View General-Purpose 3D Object Detection

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

How to project 2d feature to 3d volume?

VLadImirluren opened this issue · comments

I am a rookie of 3d perception. Recently I start to be a reseacher in this field.

Can you tell me how you project the 2d feature to 3d volume? I can't understand, it is a ill-posed problem, isn't it?

image

With known extrinsic and intrinsic camera matrices, there is a correspondence between a 3D point and a 2D pixel in the image. With such mapping several 3D points correspond to one 2D pixel.

yeah, that's what I wondering about... So what's the corelation between the 3D voxel volume with 2d feature?

I mean several 3d points correspond to one 3d pixel so how did you project 2d feature to 3d volume?

According to the corresponding code, I think you extract the 3d point cloud then put them in 2d voxel volume, right?(I guess)

image

Thanks to your reply~

Best wishes~

Sure, more formulas about this backproject function are given in section 3.1 of our paper.

Sure, more formulas about this backproject function are given in section 3.1 of our paper.

Thanks !Get it ~