Search papers, labs, and topics across Lattice.
This paper introduces ARDepth, a novel approach to monocular depth estimation that leverages structured auto-regressive generation rather than relying on global refinement. By incorporating Scale-Progressive Conditioning (SPC) and Semantic-Aware Guidance (SAG), ARDepth progressively constructs depth representations, allowing for the capture of fine-grained local details while maintaining coherent global geometry. Empirical results indicate that ARDepth outperforms existing methods, demonstrating the effectiveness of auto-regressive generation in producing structurally consistent depth predictions across varying spatial scales.
ARDepth reveals that structured auto-regressive generation can significantly enhance monocular depth estimation by capturing local details without sacrificing global coherence.
Diffusion models have recently become the dominant paradigm for monocular depth estimation (MDE). However, they implicitly assume that depth can be recovered as a globally smooth field through iterative denoising, which does not explicitly reflect the piecewise and scale-dependent organization of scene geometry. In practice, geometric structure emerges progressively across spatial scales, where coarse layout, surfaces, and boundaries are constructed in a hierarchical manner. Motivated by this observation, we introduce ARDepth, which formulates depth estimation as structured auto-regressive generation. Instead of recovering depth through global refinement, ARDepth progressively constructs depth representations as spatial resolution increases. To support this generative process, we introduce Scale-Progressive Conditioning (SPC) to inject multi-scale visual features at each generation stage, and Semantic-Aware Guidance (SAG) to provide scene-level semantic priors that enhance global structural consistency. Together, these designs enable the model to capture fine-grained local details while maintaining coherent global geometry. Empirical results demonstrate that our approach achieves strong performance and produces structurally consistent depth predictions across scales, validating auto-regressive generation as a promising alternative paradigm for geometric modeling.
