OceanPile: A Large-Scale Multimodal Ocean Corpus for Foundation Models

TL;DR

OceanPile introduces a comprehensive multimodal ocean dataset to enhance foundation models in marine AI, addressing data fragmentation and alignment challenges with high-quality, diverse data sources.

Contribution

The paper presents OceanPile, a large-scale, multimodal ocean dataset with a novel pipeline for data synthesis, quality control, and evaluation, tailored for marine artificial intelligence.

Findings

Models trained on OceanPile show significant performance improvements.

The dataset enables better semantic alignment across ocean data modalities.

OceanPile is publicly available to foster marine AI research.

Abstract

The vast and underexplored ocean plays a critical role in regulating global climate and supporting marine biodiversity, yet artificial intelligence has so far delivered limited impact in this domain due to a fundamental data bottleneck. Specifically, ocean data are highly fragmented across disparate sources and inherently exhibit multi-modal, high-noise, and weakly labeled characteristics, lacking unified schemas and semantic alignment. Although Multimodal Large Language Models (MLLMs) have achieved remarkable success in general domains, their application to ocean science remains severely constrained by the absence of large-scale, well-aligned multimodal datasets tailored to marine environments. To bridge this gap, we introduce OceanPile, a large-scale multimodal corpus designed for ocean foundation models. It comprises three key components: OceanCorpus, a unified collection integrating sonar data, underwater imagery, marine science visuals, and scientific text from diverse authoritative sources; OceanInstruction, a high-quality instruction dataset synthesized via a novel pipeline guided by a hierarchical Ocean Concept Knowledge Graph; and OceanBenchmark, a manually curated evaluation benchmark for rigorous assessment. We establish a multi-stage quality control process to ensure scientific validity and alignment across modalities. Experimental validation demonstrates significant performance improvements for models trained on our data. All datasets are publicly released to advance the field of marine artificial intelligence and empower domain-specific MLLMs.