Tokenize Anything via Prompting

Ting Pan, Lulu Tang, Xinlong Wang +1

We present a unified, promptable model capable of simultaneously segmenting, recognizing, and captioning anything. Unlike SAM, we aim to build a versatile region representation in…