High-Resolution Underwater Camouflaged Object Detection: GBU-UCOD Dataset and Topology-Aware and Frequency-Decoupled Networks

TL;DR

This work tackles deep-sea UCOD by introducing DeepTopo-Net, a physically grounded framework that preserves topology under non-uniform optical degradation. It jointly leverages a Water-Conditioned Adaptive Perceptor (WCAP) with a learnable Riemannian metric and an Abyssal-Topology Refinement Module (ATRM) to stabilize fragile, slender structures, integrated within a MAE-ViT backbone and trained with a dual-task loss. The authors also present GBU-UCOD, a high-resolution 2K dataset covering marine vertical zonation, to enable robust evaluation across depth, including abyssal and hadal zones. Empirical results on MAS3K, RMAS, and GBU-UCOD show state-of-the-art performance, with significant gains in structural integrity and skeletal connectivity, and ablations confirm the complementary benefits of WCAP and ATRM. The approach offers a practical advance for autonomous underwater perception, with public release of datasets and code and potential extensions to multi-modal fusion for even greater resilience in complex seabed environments.

Abstract

Underwater Camouflaged Object Detection (UCOD) is a challenging task due to the extreme visual similarity between targets and backgrounds across varying marine depths. Existing methods often struggle with topological fragmentation of slender creatures in the deep sea and the subtle feature extraction of transparent organisms. In this paper, we propose DeepTopo-Net, a novel framework that integrates topology-aware modeling with frequency-decoupled perception. To address physical degradation, we design the Water-Conditioned Adaptive Perceptor (WCAP), which employs Riemannian metric tensors to dynamically deform convolutional sampling fields. Furthermore, the Abyssal-Topology Refinement Module (ATRM) is developed to maintain the structural connectivity of spindly targets through skeletal priors. Specifically, we first introduce GBU-UCOD, the first high-resolution (2K) benchmark tailored for marine vertical zonation, filling the data gap for hadal and abyssal zones. Extensive experiments on MAS3K, RMAS, and our proposed GBU-UCOD datasets demonstrate that DeepTopo-Net achieves state-of-the-art performance, particularly in preserving the morphological integrity of complex underwater patterns. The datasets and codes will be released at https://github.com/Wuwenji18/GBU-UCOD.

Figures (4)

• Figure 1: Motivation.(a) Challenge: Artificial spotlights obscure fine topology (e.g., tentacles) in dark zones. (b) Baseline Failure: Rigid sampling grids miss these faint features, causing fragmentation. (c) Our Success: Our DeepTopo-Net recovers the complete structure using WCAP (warped sampling) and ATRM (topology stitching).
• Figure 2: Vertical zonation overview illustrating the challenges across different depth strata in the proposed GBU-UCOD benchmark.
• Figure 3: The overall architecture of DeepTopo-Net. The framework utilizes a MAE-ViT backbone for semantic reconstruction. The WCAP module compensates for physical degradation using Riemannian metric warping, while the ATRM utilizes directional geometric mining and skeletal constraints to ensure topological connectivity of slender targets.
• Figure 4: Qualitative comparison of segmentation results on challenging scenarios from the GBU-UCOD dataset. From left to right: Input Image, Ground Truth (GT), DeepTopo-Net (Ours), SAM2-Unet xiong2026sam2, MAS-SAM wang2024massam, H2Former he2023h2former, and MASNet 10113781. Our method consistently preserves the topological connectivity of slender appendages (top rows) and accurately segments transparent bodies (bottom rows) where other SOTA methods fail or produce fragmented masks.