Expose Camouflage in the Water: Underwater Camouflaged Instance Segmentation and Dataset

TL;DR

This work tackles underwater camouflaged instance segmentation by introducing UCIS4K, the first dataset dedicated to camouflaged marine organisms, and UCIS-SAM, a SAM-based model augmented with three domain-specific modules. The Channel Balance Optimization Module (CBOM) mitigates underwater color distortion, the Frequency Domain True Integration Module (FDTIM) emphasizes intrinsic object features, and the Multi-scale Feature Frequency Aggregation Module (MFFAM) sharpens boundaries across frequency bands to guide SAM’s segmentation. Extensive experiments across UCIS4K, UIIS, COD10K, and NC4K show that UCIS-SAM achieves state-of-the-art performance, with significant improvements in $AP$, $AP_{50}$, and $AP_{75}$ compared to existing methods, and ablation studies confirm the critical contributions of CBOM, FDTIM, and MFFAM. This work provides both a new underwater dataset and a domain-adapted segmentation framework that can enhance ecological monitoring and underwater exploration tasks by delivering more accurate camouflaged object delineation under challenging lighting and turbidity conditions.

Abstract

With the development of underwater exploration and marine protection, underwater vision tasks are widespread. Due to the degraded underwater environment, characterized by color distortion, low contrast, and blurring, camouflaged instance segmentation (CIS) faces greater challenges in accurately segmenting objects that blend closely with their surroundings. Traditional camouflaged instance segmentation methods, trained on terrestrial-dominated datasets with limited underwater samples, may exhibit inadequate performance in underwater scenes. To address these issues, we introduce the first underwater camouflaged instance segmentation (UCIS) dataset, abbreviated as UCIS4K, which comprises 3,953 images of camouflaged marine organisms with instance-level annotations. In addition, we propose an Underwater Camouflaged Instance Segmentation network based on Segment Anything Model (UCIS-SAM). Our UCIS-SAM includes three key modules. First, the Channel Balance Optimization Module (CBOM) enhances channel characteristics to improve underwater feature learning, effectively addressing the model's limited understanding of underwater environments. Second, the Frequency Domain True Integration Module (FDTIM) is proposed to emphasize intrinsic object features and reduce interference from camouflage patterns, enhancing the segmentation performance of camouflaged objects blending with their surroundings. Finally, the Multi-scale Feature Frequency Aggregation Module (MFFAM) is designed to strengthen the boundaries of low-contrast camouflaged instances across multiple frequency bands, improving the model's ability to achieve more precise segmentation of camouflaged objects. Extensive experiments on the proposed UCIS4K and public benchmarks show that our UCIS-SAM outperforms state-of-the-art approaches.

Figures (13)

• Figure 1: A comparative analysis of our dataset and method against existing datasets and methods. (a) The proportion of underwater images in the our UCIS4K, COD10Kfan2020camouflaged, and NC4Klv2021simultaneously. (b) Comparison of segmentation results. The CIS models OSFormer pei2022osformer and DCNet luo2023camouflaged confuse the instance with underwater surroundings, while ours can segment it more accurately.
• Figure 2: Examples of various challenging attributes from our UCIS4K dataset. It includes camouflaged objects with similar colors and textures to the background, blurred contours, small sizes, multiple objects, occlusion, complex contours, transparency, and underwater scenes with light and shadow effects.
• Figure 3: Examples of uncamouflaged objects and camouflaged objects. (a) Uncamouflaged objects appear unclear due to motion or backlighting. (b) Color camouflaged objects, (c) Texture camouflaged objects, (d) Edge blur camouflaged objects.
• Figure 4: The resolution distribution of images in the camouflaged dataset. (a) UCIS4K, (b) COD10Kfan2020camouflaged, (c) NC4Klv2021simultaneously. Our UCIS4K dataset contains higher-resolution images than both COD10K and NC4K, providing richer visual information.
• Figure 5: The distribution of the number of camouflaged instances per image in the UCIS4K, COD10Kfan2020camouflaged, and NC4Klv2021simultaneously dataset.