search

CVMI-Lab / PLA

(CVPR 2023) PLA: Language-Driven Open-Vocabulary 3D Scene Understanding & (CVPR2024) RegionPLC: Regional Point-Language Contrastive Learning for Open-World 3D Scene Understanding
Apache License 2.0
262 stars 11 forks source link

About instance segmentation #15

Closed ZheningHuang closed 1 year ago

ZheningHuang commented 1 year ago

Hi, Thank you so much for the great work. I am writing to clarify a question regarding the instance segmentation:

I can see that you have modified the SoftGroup architecture by replacing anything that revealed novel category class labels. My question is if the ground truth (GT) class-agnostic instance mask is needed during training. My understanding is yes, because you need that information to train the Class-agnostic Score Head as well as the Offset Head, both of which require GT class-agnostic mask information. Please correct me if I am wrong. Thank you so much for your time Best, Zhening

Dingry commented 1 year ago

Hi, Zhening, gt instance label is required only for base categories. We do not use any gt label for novel categories during training.