search

Pointcept / SegmentAnything3D

[ICCV'23 Workshop] SAM3D: Segment Anything in 3D Scenes
https://arxiv.org/abs/2306.03908
MIT License
1.01k stars 70 forks source link

How to Use group_ids in Final .pth Files and Link to Semantic Labels #55

Open JiweiHou opened 1 week ago

JiweiHou commented 1 week ago

Thanks for your open source, very nice work。 I have a small questiion. In the final step of sam3d.py, we generated .pth files containing group_ids for each scene, such as Scene0000_0.pth for scene Scene0000_0. However, the IDs in these .pth files do not directly correspond to the IDs in the scannetv2-labels.combined.tsv file, which would allow us to obtain the desired semantic information, like the point clouds for tables, chairs, floors, and other objects. Additionally, once we have generated a file like Scene0000_0.pth, what exactly is its intended use? Is it primarily for visualization in combination with the coordinates? Could you explain this more, because I'm wondering how I can get a point cloud with semantic information using SAM. Thank you, looking forward to your reply.

yhyang-myron commented 1 week ago

Hi, thanks for your interest in our work! The predicted group_id should be used with the processed 3D point cloud data. Their sequences are corresponding.