🤖 AI Summary
Addressing challenges in underwater monocular depth and surface normal estimation (MDSNE)—including difficulty modeling transparent/reflective surfaces, sensitivity to real-world noise, and poor generalization from synthetic data—this paper proposes a lightweight CNN-Transformer hybrid architecture. We introduce the Depth-Normal Evaluation and Selection Algorithm (DNESA), the first domain-aware pseudo-label quality assessment and filtering framework for underwater vision. Our method integrates multi-model ensemble pseudo-label generation, curriculum-based knowledge distillation, and underwater-specific evaluation metrics to enhance generalization and robustness. Evaluated on multiple underwater benchmarks, our approach achieves state-of-the-art accuracy while reducing model parameters by 90% and training cost by 80%. The compact design enables real-time inference on low-power embedded devices and has been successfully deployed on underwater robots and autonomous underwater vehicles (AUVs).
📝 Abstract
Monocular Depth and Surface Normals Estimation (MDSNE) is crucial for tasks such as 3D reconstruction, autonomous navigation, and underwater exploration. Current methods rely either on discriminative models, which struggle with transparent or reflective surfaces, or generative models, which, while accurate, are computationally expensive. This paper presents a novel deep learning model for MDSNE, specifically tailored for underwater environments, using a hybrid architecture that integrates Convolutional Neural Networks (CNNs) with Transformers, leveraging the strengths of both approaches. Training effective MDSNE models is often hampered by noisy real-world datasets and the limited generalization of synthetic datasets. To address this, we generate pseudo-labeled real data using multiple pre-trained MDSNE models. To ensure the quality of this data, we propose the Depth Normal Evaluation and Selection Algorithm (DNESA), which evaluates and selects the most reliable pseudo-labeled samples using domain-specific metrics. A lightweight student model is then trained on this curated dataset. Our model reduces parameters by 90% and training costs by 80%, allowing real-time 3D perception on resource-constrained devices. Key contributions include: a novel and efficient MDSNE model, the DNESA algorithm, a domain-specific data pipeline, and a focus on real-time performance and scalability. Designed for real-world underwater applications, our model facilitates low-cost deployments in underwater robots and autonomous vehicles, bridging the gap between research and practical implementation.