Dataset for Hierarchical Object Detection of Nordic Fish Species

Published: 18-05-2020| Version 1 | DOI: 10.17632/b4kcw9r32n.1
Contributors:
Espen Kalhagen,
Ørjan Olsen

Description

The dataset contains data used in the thesis "Hierarchical Fish Species Detection in Real-Time Video Using YOLO". It consists of the weights of the models that was used for the different experiments, some framework configurations, and the object detection dataset. The detector runs on the Darknet framework. The object detection dataset is a collection of 1879 images of underwater fish taken from a stationary camera in Lindesnes, Norway. Each image has a corresponding annotation file that defines a bounding box and class for each fish. The images are annotated with hierarchical classification and biological taxonomy in mind. The hierarchy is defined in fish_taxonomy.xml. This means that if the species cannot be discerned, a higher class in the hierarchy will be used. There is 7721 annotated fish in the dataset.

Files