Live page · Day archive

GroundAnything: Reconciling Parallel Decoding with Precise Visual Grounding at Flash Speed

Read the original at HF Daily Papers

Summary

GroundAnything is a 4B-parameter model that uses bidirectional diffusion to enable parallel decoding for precise visual grounding, achieving a 72.42% score across 30 benchmarks.

Carried by: HF Daily Papers. First seen: .