How to set some semantic labels to void

Harry-Zhi / semantic_nerf

The implementation of "In-Place Scene Labelling and Understanding with Implicit Scene Representation" [ICCV 2021].

Other

418 stars 54 forks source link

How to set some semantic labels to void #12

Closed jmwang0117 closed 2 years ago

jmwang0117 commented 2 years ago

Thank you for your awesome work！！！

I plan to set the semantics of some instances to void when training the model, because I don't want some instances to be segmented, what can I do?

thanks！！！

Harry-Zhi commented 2 years ago

Hi @jmwang0117,

As far as I can well, the most straightforward way could be setting the instances that you did not want to predict as label 0 (void class), so they will not be penalised during training. Though at inference time, you will still predict some results on these regions and you could simply mask them out for visualisation purposes. See issue #3 .