Quality-Driven and Diversity-Aware Sample Expansion for Robust Marine Obstacle Segmentation

2025-12-15

AI Summary

To address the robustness challenge of marine obstacle segmentation under low-quality imaging conditions—such as sun glint, haze, and dynamic wave interference—this paper proposes an inference-time sample augmentation method that requires no retraining of diffusion models. To overcome the limited output diversity of existing mask-conditioned diffusion models under low-entropy prompts, we construct a class-aware high-entropy style library to generate semantically rich prompts. Furthermore, we introduce Contrastive Output Divergence (COD)-guided proportional control and an adaptive annealing sampling mechanism, which jointly enhance output diversity while preserving structural fidelity. Evaluated across multiple marine obstacle segmentation benchmarks, our method consistently improves the performance of mainstream segmentation models—particularly boosting accuracy and visual diversity for rare and texture-sensitive classes. This work establishes a novel paradigm for robust segmentation in few-shot, low-quality imaging scenarios.

Technology Category

• Computer Vision: Segmentation
• Machine Learning: Adversarial Learning & Robustness
• Search and Optimization: Sampling/Simulation-based Search

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking

Abstract

Marine obstacle detection demands robust segmentation under challenging conditions, such as sun glitter, fog, and rapidly changing wave patterns. These factors degrade image quality, while the scarcity and structural repetition of marine datasets limit the diversity of available training data. Although mask-conditioned diffusion models can synthesize layout-aligned samples, they often produce low-diversity outputs when conditioned on low-entropy masks and prompts, limiting their utility for improving robustness. In this paper, we propose a quality-driven and diversity-aware sample expansion pipeline that generates training data entirely at inference time, without retraining the diffusion model. The framework combines two key components:(i) a class-aware style bank that constructs high-entropy, semantically grounded prompts, and (ii) an adaptive annealing sampler that perturbs early conditioning, while a COD-guided proportional controller regulates this perturbation to boost diversity without compromising layout fidelity. Across marine obstacle benchmarks, augmenting training data with these controlled synthetic samples consistently improves segmentation performance across multiple backbones and increases visual variation in rare and texture-sensitive classes.

Problem

• Enhances marine obstacle segmentation under challenging conditions like sun glitter and fog.
• Addresses limited diversity in marine datasets due to scarcity and structural repetition.
• Improves synthetic data diversity without retraining diffusion models for better robustness.

Innovation

• Class-aware style bank for high-entropy prompts
• Adaptive annealing sampler perturbs early conditioning
• COD-guided controller boosts diversity without layout loss