BARIS: Boundary-Aware Refinement with Environmental Degradation Priors for Robust Underwater Instance Segmentation

TL;DR

This work tackles underwater instance segmentation under challenging visual distortions such as attenuation, scattering, and color shifts. It introduces BARIS-Decoder for boundary-focused multi-scale refinement and ERA-tuning to model underwater degradations with substantial parameter efficiency, complemented by the Boundary-Aware Cross-Entropy (BACE) loss that leverages range-null space decomposition to sharpen object boundaries. The approach achieves state-of-the-art results on UIIS and USIS10K, with notable gains in mAP and boundary precision while using a fraction of trainable parameters compared to full fine-tuning. The proposed framework offers a robust, efficient solution for real-world underwater perception, enabling better autonomous operation of underwater robots and systems across diverse environments.

Abstract

Underwater instance segmentation is challenging due to adverse visual conditions such as light attenuation, scattering, and color distortion, which degrade model performance. In this work, we propose BARIS-Decoder (Boundary-Aware Refinement Decoder for Instance Segmentation), a framework that enhances segmentation accuracy through feature refinement. To address underwater degradations, we introduce the Environmental Robust Adapter (ERA), which efficiently models underwater degradation patterns while reducing trainable parameters by over 90\% compared to full fine-tuning. The integration of BARIS-Decoder with ERA-tuning, referred to as BARIS-ERA, achieves state-of-the-art performance, surpassing Mask R-CNN by 3.4 mAP with a Swin-B backbone and 3.8 mAP with ConvNeXt V2. Our findings demonstrate the effectiveness of BARIS-ERA in advancing underwater instance segmentation, providing a robust and efficient solution.

Figures (9)

• Figure 1: Comparison of our approach with state-of-the-art methods on the UIIS dataset. USIS-SAM lian2024diving uses a ViT-H backbone, while all other methods adopt Swin-B. Our BARIS-ERA method achieves the best performance across all AP metrics.
• Figure 2: The Environmental Robust Adapter (ERA) integrated into a Swin Transformer block. ERA is positioned at the end of each block, while the rest of the network remains frozen. This design enables efficient adaptation to underwater distortions without modifying the core model architecture.
• Figure 3: The architecture of the proposed BARIS-Decoder for underwater image instance segmentation. BARIS-Decoder consists of (a) Multi-Stage Gated Refinement Network (defined in Section \ref{['MSGRN']}) and (b) Depthwise Separable Upsample (defined in Section \ref{['DSUpsample']}).
• Figure 4: The architecture of the Environmental Robust Adapter (ERA). ERA enhances feature representations through multi-scale feature extraction (MSFE) and environmental adaptation.
• Figure 5: Qualitative comparison with state-of-the-art methods on the UIIS Dataset, using Swin Transformer (left two images) and ConvNeXt V2 (right two images) backbones.