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

Question about binary head #17

Closed SxJyJay closed 1 year ago

SxJyJay commented 1 year ago

Thanks for your great work and high-quality open-source code! I have some questions regarding the paper. Since the open vocabulary problem can only access the annotation of the base categories during training, how are the ground-truth labels (i.e., y^{b} in Eq(4) of the main paper) indicating the base and novel categories set when calculating the binary loss (Eq(4) of the main paper)? Theoretically, any information about the novel class should be unseen during the training.

jihanyang commented 1 year ago

Hello, we regard all unannotated points other than base-categories as the novel categories points. We think it will be better to recognize this as unseen category points

SxJyJay commented 1 year ago

Thanks for your timely reply! My question is solved.