search

RozDavid / LanguageGroundedSemseg

Implementation for ECCV 2022 paper Language-Grounded Indoor 3D Semantic Segmentation in the Wild
98 stars 14 forks source link

How to do the instance augmentation? #16

Closed MurphyJUAN closed 1 year ago

MurphyJUAN commented 1 year ago

Hi, so many thanks for your wonderful work! When I want to do the instance augmentation mentioned in paper section 3.2, I set the --instance_augmentation to be 'raw' in config.py and use focal loss. However, I get the following error when I run the code:

image

It seems instance augmentation would change the shape of the label. How can I fix it? Thanks.

RozDavid commented 1 year ago

Hey again,

The flag that you are looking for for tail instance sampling is sample_tail_instances, where you should also set the path where your instance meshes are saved. Please check the implementation for more details here.

What you set to True is an other flag where we tested language-based augmentations with object properties and adjectives, but didn't follow up for this paper. You can leave that unchanged or experiment with it if it could be helpful for your use case.

Hope this helps, David