is the model can segment all types of classes?

[Yaoxingtian]

Hi,thanks for your great work!
it sees x-decoder is zero-shot segment model, but when i test it by my data which including floor and bed class , it can not work well, could you pls. explain it ? thanks!

[jwyang]

Hi, @Yaoxingtian,

thanks for rasing this question. Could you elaborate a bit more on how you use our x-decoder mode?

I just tried on image and it works well:

[Yaoxingtian]

I tried all demo script in X-decoder, and in segment demo, the stuff_classes just like: ['zebra','antelope','giraffe','ostrich','sky','water','grass','sand','tree'], there is no floor or sofa class, could you tell me how do you do that ?
what i did is:
python demo/demo_semseg.py evaluate --conf_files configs/xdecoder/svlp_focalt_lang.yaml --overrides WEIGHT ./weights/xdecoder_focalt_best_openseg.pt

[Yaoxingtian]

[jwyang]

You need to replace ['zebra','antelope','giraffe','ostrich','sky','water','grass','sand','tree'] with your own list of categories.

[Yaoxingtian]

You need to replace ['zebra','antelope','giraffe','ostrich','sky','water','grass','sand','tree'] with your own list of categories.

you are right, now I know how to do that !
How many class does xdecoder segment ?
And ,can i use Open Vocabulary Segmentation:python eval.py to get this result too ?

[jwyang]

You can segment an arbitrary number of categories, but please make sure you replace the categories with your own.