NeMal: Never ending Marine Learning - Unleashing the Power of Controllable Image Synthesis for Promoting Marine Visual Understanding
Keywords: never ending marine learning, controllable image synthesis, foundation models, vision language analysis
Abstract: The relentless pursuit of marine learning is required by the essential need to understand and protect the complex marine ecosystems that cover over 70% of the surface of our planet. Due to the specific underwater/marine environments, the data collection and labeling are expensive and labor-intensive, also limited to user groups with special equipment. Existing marine visual learning just optimizes models from a small set of marine data with human labels, which cannot fit the essence of ongoing marine exploration. In this work, we propose NeMal, a \underline{N}ever-\underline{e}nding \underline{Ma}rine \underline{L}earning system that harnesses controllable image synthesis and efficient foundation models to perform never-ending marine visual synthesis and understanding. Based on NeMal, we produce MarineSynth, which is the first large-scale marine synthetic dataset to date, featuring more than 4 million unique text prompts and corresponding text-to-image outputs with pseudo labels from text prompts or foundation models. The experiments on downstream classification, segmentation, and vision-language understanding tasks demonstrate the promise of utilizing synthetic data to promote marine visual understanding, significantly reducing human efforts in both data collection and labeling.
Primary Area: applications to physical sciences (physics, chemistry, biology, etc.)
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/2025/AuthorGuide.
Reciprocal Reviewing: I understand the reciprocal reviewing requirement as described on https://iclr.cc/Conferences/2025/CallForPapers. If none of the authors are registered as a reviewer, it may result in a desk rejection at the discretion of the program chairs. To request an exception, please complete this form at https://forms.gle/Huojr6VjkFxiQsUp6.
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: 1514
Loading