Training-Free Target-Aware Refinement of 3D Gaussian Language Map

Key points

• We propose a training-free, target-aware refinement of an already-distilled language map.1
• Guided by a text-promptable segmenter, the visual embedding inside each detected target region is blended with the target text embedding, a convex combination whose coefficient sets the amount of injected prior, while non-target regions are left untouched.1
• We further identify a failure mode of injection in language-goal localization.1
• It also transfers as a plug-in to off-the-shelf field-based and object-level language maps and supports a real-world demonstration in which a legged robot navigates to natural-language goals.1

Sources (1)

• [1]Training-Free Target-Aware Refinement of 3D Gaussian Language Map
  arXiv (AI, ML, NLP, CV, robotics, multi-agent) · Oct 15, 10:00 AM

  Distilling vision-language features into 3D Gaussian Splatting yields a language map that can be queried with arbitrary text.

  We propose a training-free, target-aware refinement of an already-distilled language map.

Extractive summary: sentences quoted from the sources.