AquaStereo: Enabling Underwater Stereo Matching via Depth-Conditioned Diffusion and Geometry Self-Distillation

2026-07-05

AI Summary

This work addresses the challenges of underwater stereo matching, which suffers from severe performance degradation due to scarce real-world data and image quality deterioration that hinder reliable feature correspondence. To overcome these limitations, the authors propose AquaStereo, a novel framework that first leverages a depth-conditioned diffusion model to synthesize geometrically consistent underwater stereo image pairs. It then employs a geometric self-distillation mechanism, where a teacher model pretrained on terrestrial data guides a student model to acquire robust matching capabilities. Furthermore, AquaStereo introduces a learnable perceptual frame that fuses a video backbone with an image encoder, enhanced by left-right consistency constraints, stage-weighted sequential loss, and shared pseudo-target supervision. This integrated approach effectively mitigates data scarcity and feature degradation, significantly improving both robustness and zero-shot generalization across diverse and challenging underwater environments.

Abstract

Learning-based stereo matching models struggle in underwater environments due to scarce in-domain data and the difficulty of extracting discriminative correspondences from degraded imagery. In this work, we present $\textbf{AquaStereo}$, a perception-enhanced framework with a data simulation pipeline and a self-distillation strategy that jointly address data scarcity and feature degradation in underwater stereo matching. First, a depth-conditioned diffusion pipeline renders underwater stereo pairs while preserving binocular geometry, with a lightweight left-right consistency module ensuring geometric alignment. Training on this synthetic corpus effectively narrows the terrestrial-underwater gap and improves zero-shot robustness. Second, a frozen binocular teacher trained on clean terrestrial pairs guides a student exposed to rendered underwater pairs with perturbations. A stage-weighted sequence loss is performed to align the student's disparities with the teacher's geometry, while a clean-branch supervision with shared pseudo targets prevents scale drift. To further enhance feature stability under turbidity and low texture, we introduce learnable perception frames, a perception-enhanced feature formulation that constructs robust matching descriptors by fusing temporal cues from two auxiliary views encoded by a video backbone with semantic features extracted by a strong image encoder. Extensive experiments demonstrate that $\textbf{AquaStereo}$ substantially improves robustness and zero-shot generalization in challenging underwater scenarios. The code is available at https://github.com/qz-wei/AquaStereo.

Problem

• underwater stereo matching
• data scarcity
• feature degradation
• degraded imagery
• correspondence extraction

Innovation

• underwater stereo matching
• depth-conditioned diffusion
• geometry self-distillation
• learnable perception frames
• zero-shot generalization