Answers are grounded in this paper and cite their sources.
Fish play a vital role in the ecosystems of rivers and lakes, and the application of deep learning-based intelligent identification methods offers an efficient and accurate approach to water environment regulation. However, the random spatial distribution of underwater fish complicates the collection of extensive real-image datasets. This scarcity of images diminishes the generalization capability of deep learning models, thereby limiting their practical applicability. To address this challenge, we propose a diffusion model-based approach that involves pre-training on an open-source underwater fish dataset and subsequently generating realistic, diverse underwater fish images from pure noise. The generated images achieved 6.8070, 9.1829 and 26.3132 on the evaluation metrics NIQE, PIQE and BRISQUE, respectively.
No takes yet. Share an insight, caveat, or question.
Xinyu Wang (2025) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: