Is it possible we can define our own labels?

[PoopBear1]

Thanks for sharing this great work!

Can we define our labels or apply transfer learning to this project? I did not figure out how do we run this project if our dataset is not from SA-B dataset. e.g. I want to apply it on an indoor scene where the labels are mainly the furniture. Would simple addition of config files on configs work?

Appreciate your help!

[Jiaqi-Chen-00]

Simply modifying the configuration file is not enough. We have open-sourced a close-set version of SSA (https://github.com/fudan-zvg/Semantic-Segment-Anything/blob/main/scripts/pipeline.py#L104), which supports using your own segmentation model with the desired segmentation categories as the semantic branch to provide category predictions. You can use this to achieve your specified category requirements. Please refer to the Quick Start in the README to try to run SSA and make some modifications.