OS-Ship-1K: A CycleGAN-Based Optical-SAR Multimodal Ship Detection Dataset

lingjie Jiang, Xin Luo, Yuhao Lin, Kaiyuan Xu, Haewoon Nam, Dongliang Peng

Keywords: Multimodal, CycleGAN, Ship Detection, Dataset

TL;DR: This paper introduces the first publicly available dataset for optical–SAR multimodal ship detection, termed OS-Ship-1K, based on CycleGAN.

Abstract: Over the past decade, multimodal remote sensing image fusion techniques have developed rapidly. By integrating multi-source remote sensing data, it is possible to obtain more comprehensive and accurate observational information while compensating for the limitations of single-modality imagery. However, in the field of optical–synthetic aperture radar (SAR) multimodal ship detection, challenges in achieving spatiotemporal consistency during data acquisition, coupled with constraints related to data privacy and national defense security, have resulted in a scarcity of publicly available datasets, severely hindering technological advances. To address this issue, this paper introduces the first publicly available dataset for optical–SAR multimodal ship detection, termed OS-Ship-1K, based on CycleGAN. The dataset comprises 1,000 pairs of aligned optical and SAR ship images, annotated with two categories: \textit{Ship} and \textit{Ships}. OS-Ship-1K covers both inshore and offshore scenarios while meeting the detection requirements for both sparse and dense targets. Furthermore, we conduct a comprehensive evaluation of 14 single-modal detectors and 6 multimodal fusion detectors on OS-Ship-1K to establish baseline standards. We hope that the release of the OS-Ship-1K dataset will attract broader attention and engagement from the research community, thereby driving new breakthroughs in optical–SAR multimodal ship detection. The dataset will be released after acceptance.

Primary Area: datasets and benchmarks

Code Of Ethics: I acknowledge that I and all co-authors of this work have read and commit to adhering to the ICLR Code of Ethics.

Submission Guidelines: I certify that this submission complies with the submission instructions as described on https://iclr.cc/Conferences/2026/AuthorGuide.

Anonymous Url: I certify that there is no URL (e.g., github page) that could be used to find authors’ identity.

No Acknowledgement Section: I certify that there is no acknowledgement section in this submission for double blind review.

Submission Number: 12578