New MARINER benchmark tests AI models on fine-grained maritime understanding with 16,629 images across 63 vessel types and complex ocean scenarios.

MARINER introduces a novel Entity-Environment-Event (3E) framework to evaluate AI performance on maritime perception and reasoning tasks

Dataset contains 16,629 multi-source maritime images covering 63 fine-grained vessel categories, adverse environmental conditions, and 5 dynamic maritime incidents

Evaluations reveal that advanced Multimodal Large Language Models struggle with fine-grained discrimination and causal reasoning in complex marine scenes

Benchmark supports multiple tasks including fine-grained classification, object detection, and visual question answering for maritime applications