Semantic segmentation example

[crapthings]

How to use PE-Spatial-G14-448 to get binary mask?

a photo of an apple

input apple, then get the mask?

[dbolya]

Both PE core and PE lang are feature extractors. They output features useful for downstream tasks, but they themselves don't do any tasks like zero-shot semantic segmentation. You'll either have to train a semantic segmentation head yourself or plug PE spatial into a zero-shot method (see #20).