SceneBind: Binding What and Where Across Vision, Audio and Language

By Mingfei Chen · Paper · cs.CV

We present SceneBind, an omni-modal representation of realistic scenes with joint semantic and 3D spatial understanding across vision, audio and language. Existing omni-modal encoders excel at instance-level semantics (i.e., what is present), but often lack explicit spatial struc

Cs.cv

View original

HomeResourceLoading…