Reproduce “NAUTILUS: A Large Multimodal Model for Underwater Scene Understanding” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Wei Xu, Cheng Wang, Dingkang Liang, Zongchuang Zhao, Xingyu Jiang, Peng Zhang, Xiang Bai
Researchers at Huazhong University of Science and Technology and National University of Defense Technology developed NAUTILUS, a large multimodal model that performs eight diverse underwater scene understanding tasks. It incorporates a physically-motivated feature enhancement module and a new multi-task dataset, demonstrating robust performance across varied degraded conditions.
Key insights
- Directly applying pixel-level underwater image enhancement methods can lead to information loss that degrades performance on high-level semantic understanding tasks.
- Integrating physical priors from underwater imaging into a feature-space enhancement module (VFE) effectively restores degraded visual information while preserving essential semantic cues.
- Training with a comprehensive, multi-task, and multi-granular instruction-following dataset is critical for adapting LMMs to achieve robust and versatile performance in challenging, domain-specific environments like underwater scenes.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.