In contrast, the existing 3D visual grounding (left) relies on human reasoning and references for detection. The illustration clearly distinguishes that observation and reasoning are manually executed ...