search

xinge008 / Cylinder3D

Rank 1st in the leaderboard of SemanticKITTI semantic segmentation (both single-scan and multi-scan) (Nov. 2020) (CVPR2021 Oral)
Apache License 2.0
859 stars 180 forks source link

Cannot find point-wise prediction and loss function. #42

Closed mandal4 closed 3 years ago

mandal4 commented 3 years ago

Thanks for sharing your kind code. Unfortunately i couldn't find out the code about point-wise prediction and corresponding loss function. I found out the output of model is [#bs, 20, 480,360,32] which is a form of voxel. And the corresponding loss is calculated with voxel-wise class label.

Where can i find the lines about point-wise prediction and corresponding loss function?

Thanks,

FrancescoMandru commented 3 years ago

Maybe is not implemented yet

https://github.com/xinge008/Cylinder3D/issues/8#issuecomment-737069254

JBKnights commented 3 years ago

The point-wise predictions can be found in the validation loop - you can see that the grid is used to re-project from the voxels to the point cloud

In the current implementation there isn't a point-wise refinement module; all the losses take place purely on a voxel level.