<p>The utilization of image datasets has become a crucial aspect of fisheries research lately, revolutionizing conventional techniques and improving humans’ comprehension of aquatic ecosystems. Of all the aquatic creatures, fish is thought to be the most nutritious for the large class of nonvegetarian consumers worldwide. With more than 32,000 varieties, fish are differentiated by visual characteristics such as shape, texture, patterns, and color, which makes identification difficult for a common user. Machine learning (ML) and deep learning (DL) based models are being developed to automate fish related studies by utilizing their visual characteristics. The automated DL-based applications related to fisheries research include, but are not limited to, fish length estimation, behavioral analysis, fish detection, and classification. These applications rely on image datasets, but several challenges hinder their effectiveness. Existing datasets often suffer from limitations such as lack of defined information, uneven image quality, limited geographic coverage, and insufficient species variety. Additionally, the absence of benchmark datasets and inconsistencies in data collection, annotation accuracy, and environmental variability may affect model performance and generalizability. These gaps limit the development of robust DL-based fisheries applications. This review systematically examines the utilization of fish image datasets in fisheries research, focusing on applications, species studied, study regions, image characteristics, and associated metadata. By identifying key research gaps, the study highlights the need for improved dataset quality, standardization, and comprehensive metadata to enhance automated fisheries research. Addressing these challenges can facilitate more accurate fisheries research, monitoring, and conservation efforts, ultimately supporting sustainable aquaculture and ecosystem management.</p>