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

point wise loss #62

Closed caoyifeng001 closed 3 years ago

caoyifeng001 commented 3 years ago

thank you for releasing such a good job! why point wise loss doesn`t use lovasz loss

xinge008 commented 3 years ago

I do not try the combination of Cross-entropy and lovasz loss for point-wise loss; Not sure its performance.