SPADE: Sparsity Adaptive Depth Estimator for Zero-Shot, Real-Time, Monocular Depth Estimation in Underwater Environments

2025-10-29

AI Summary

Underwater infrastructure inspection is hindered by turbid water and complex structural geometries, while conventional manual or ROV-based approaches suffer from perceptual and operational bottlenecks. To address scale ambiguity and inefficient utilization of sparse depth priors in zero-shot monocular underwater depth estimation, this paper proposes a training-free, real-time method that enables adaptive depth densification and metric-scale recovery. Our approach uniquely integrates a pre-trained relative depth model with sparse depth priors within a two-stage framework: first performing sparse point cloud registration via a cascaded convolutional–deformable Transformer module, then refining the dense depth map using a lightweight Transformer. The method supports embedded deployment and achieves state-of-the-art accuracy and generalization on real underwater scenes, operating at over 15 FPS.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Intelligent Robots: State Estimation
• Reasoning under Uncertainty: Relational Probabilistic Models

Application Category

• Graph Algorithms and Modeling for the Web: Foundation models and LLMs for Web-related graphs
• User Modeling, Personalization and Recommendation: On-Device user modeling, personalization, and recommendation
• Responsible Web: Human-perceived consequences of algorithmic deployment on the web

Abstract

Underwater infrastructure requires frequent inspection and maintenance due to harsh marine conditions. Current reliance on human divers or remotely operated vehicles is limited by perceptual and operational challenges, especially around complex structures or in turbid water. Enhancing the spatial awareness of underwater vehicles is key to reducing piloting risks and enabling greater autonomy. To address these challenges, we present SPADE: SParsity Adaptive Depth Estimator, a monocular depth estimation pipeline that combines pre-trained relative depth estimator with sparse depth priors to produce dense, metric scale depth maps. Our two-stage approach first scales the relative depth map with the sparse depth points, then refines the final metric prediction with our proposed Cascade Conv-Deformable Transformer blocks. Our approach achieves improved accuracy and generalisation over state-of-the-art baselines and runs efficiently at over 15 FPS on embedded hardware, promising to support practical underwater inspection and intervention. This work has been submitted to IEEE Journal of Oceanic Engineering Special Issue of AUV 2026.

Problem

• Estimating dense metric depth from monocular underwater images
• Overcoming perceptual limitations in turbid water environments
• Enabling real-time depth estimation for autonomous underwater vehicles

Innovation

• Combines pre-trained depth estimator with sparse priors
• Uses two-stage scaling and refinement approach
• Runs efficiently on embedded hardware at 15 FPS