Abstract: We tackle open-world semantic segmentation, which aims at learning to segment arbitrary visual concepts in images, by using only image-text pairs without dense annotations. Existing ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback