VGLD: Visually-guided language disambiguation for monocular depth scale recovery

Monocular depth estimation can be broadly categorized into two directions: relative depth estimation, which predicts normalized or inverse depth without absolute scale, and metric depth estimation, which aims to recover depth with real-world scale. While relative methods are flexible and data-efficient, their lack of metric scale limits their utility in downstream tasks. A promising solution is to infer absolute scale from textual descriptions. However, such language-based recovery is highly sensitive to natural language ambiguity, as the same image may be described differently across perspectives and styles. To address this, we introduce VGLD (Visually-Guided Language Disambiguation), which, unlike purely language-based approaches such as RSA, leverages visual semantics to perform visually grounded language disambiguation and predict global linear transformation parameters that align relative depth with metric scale. We evaluate VGLD on representative models, including MiDaS and Depth Anything, using the NYU Depth V2 and KITTI benchmarks. Results show that VGLD effectively mitigates language-induced scale bias and improves metric depth accuracy. Furthermore, when trained on multiple datasets, VGLD serves as a lightweight universal alignment module that maintains strong performance even in zero-shot settings. Code will be released upon acceptance.