Salmon ScaleLossDataSet: Labeled Data for Fish Detection, Skin Segmentation, and Scale Loss Segmentation

This dataset contains annotated images of farmed Atlantic salmon for three computer vision tasks related to automated assessment of fish welfare: fish detection, skin segmentation, and scale loss segmentation. The primary images were captured above water in a controlled setting at a salmon aquaculture farm in Norway. Additional images were provided by NMBU. Contents - FishDetection.zip: bounding-box annotations for detecting fish - Original: 338 images with labels - AugmentedWithCropping: 676 augmented (cropped) images with labels- SkinSegmentation.zip: segmentation of the fish skin area - AllSkin: 303 images with labels - TopSkin: 240 images with labels- ScaleLossSegmentation.zip: segmentation of regions with scale loss, 167 images with labels- doc/: example images showing annotations for each task Format All annotations use the standard YOLO .txt format in an images/labels folder structure. Each image has a label file with the same name. Intended use Training and evaluating machine learning models for fish detection, skin segmentation, and scale loss quantification in aquaculture. Related publication Thomas Fjorden, Espen B. Høgstedt, Christian Schellewald, Rudolf Mester, Mette Remen, Ane Vigdisdatter Nytrø, and Annette Stahl. Image-based quantification of scale loss in fish using machine learning and computer vision. Proceedings of the 17th International Conference on Machine Vision (ICMV 2024), 2024. Please cite both the dataset and the related publication when using this data. Acknowledgements This work was supported by the Research Council of Norway through the project cAIge (344022), and by FHF – Norwegian Seafood Research Fund through the project ReduSår (901835). We thank Akvaplan-niva for providing the primary data for this research. Some images were provided by NMBU. License Creative Commons Attribution 4.0 International (CC BY 4.0)