search

Sekunde / 3D-SIS

[CVPR'19] 3D-SIS: 3D Semantic Instance Segmentation of RGB-D Scans
Other
379 stars 72 forks source link

semantic class label for chunk？ #47

Closed hfutdzx closed 3 years ago

hfutdzx commented 3 years ago

This is a wonderful work. Thank you for your contribution. This work is very helpful to me. I have a problem. I'm doing research on semantic category segmentation. I want to know how to use “datagen” code to mark semantic categories for voxels of each chunk？（20 class）

Sekunde commented 3 years ago

Hi,

Thanks for following our work. "datagen" is used for generating instance-level annotations per chunk; not per-voxel annotations. However, we do include the file Segmentation.h for per-voxel annotations in this repo, which may help you. On the other hand you can download the per-voxel annotation per chunk (20 class) processed from 3dmv. 3DMV should use the same code (Segmentation.h) for per-voxel annotation.

hfutdzx commented 3 years ago

Thanks a lot for your help！

hfutdzx commented 3 years ago

Hi，Prof.Hou，I have a question about the 3DMV train dataset.I used the 3DMV dataset to train 3DMV model，but I found that there were too many invalid samples in this dataset.more specifically，many 3D chunks have no mapping relationship with 2d images，and 3Dmv cannot be trained.