A realistic fish-habitat dataset to evaluate algorithms for underwater visual analysis

• DOI: 10.1038/s41598-020-71639-x
• Publication Date: 2020-09-04T10:03:02Z
• Authors: Alzayat Saleh; Issam H. Laradji; Dmitry A. Konovalov; Michael Bradley; David Vazquez; Marcus Sheaves
• Keywords: Benchmark (surveying); Testbed

Visual analysis of complex fish habitats is an important step towards sustainable fisheries for human consumption and environmental protection. Deep Learning methods have shown great promise scene when trained on large-scale datasets. However, current datasets tend to focus the classification task within constrained, plain environments which do not capture complexity underwater habitats. To address this limitation, we present DeepFish as a benchmark suite with dataset train test several computer vision tasks. The consists approximately 40 thousand images collected from 20 in marine-environments tropical Australia. originally contained only labels. Thus, point-level segmentation labels more comprehensive benchmark. These enable models learn automatically monitor count, identify their locations, estimate sizes. Our experiments provide in-depth characteristics, performance evaluation state-of-the-art approaches based our Although pre-trained ImageNet successfully performed benchmark, there still room improvement. Therefore, serves testbed motivate further development challenging domain vision.