Reopen Issue to Reclarify (unknown things/stuffs) : https://github.com/microsoft/X-Decoder/issues/12

Question

Reopen Issue to Reclarify (unknown things/stuffs) : https://github.com/microsoft/X-Decoder/issues/12

ffahmed opened this issue a year ago · 2 comments

I meant to say if we don't know what things and stuffs we want from a picture, I want to tell it to give me all possible things and stuffs in panoptic seg. How do I do it?

Answer 1 · 2023-02-16T00:11:20.000Z

Please go back to our paper, and other vl detection papers and understand the problem in a more detailed way. Basically we do not have a class pool, with the support of language encoder, it enables any class you want. Thanks!

Answer 2 · 2023-02-16T01:40:20.000Z

@ffahmed what you asked is more like a dense or regional captioning task. but you can enumerate as many concepts as possible in the pool to mimic this behavior.