Seaview Survey: GBR Training and Validation Dataset

Full description

This dataset consists of: (1) a series of images from a transect on Australia'a Great Barrier Reef ("images_training" and "images_validation" folders); (2) a description of the features we would like to identify within the images ("label_key.csv"); (3) a training dataset that can be used to train an interpretation algorithm to automatically identify those features in the images ("training_dataset.csv"); (4) a validation dataset that can be used to assess the performance of any interpretation algorithm ("validation_dataset.csv"). Each of the images has been processed to correct for lens distortion and cropped to standardise the area covered to approximately 1 square meter 'quadrats'. These training and validation datasets were developed by expert human interpretation of images. The images were collected on several different coral reefs within the Great Barrier Reef between 2012-2014, so represent a variety of communities and conditions. The training and validation datasets consists primarily of randomly selected points on images, but this random sample has been manually augmented to ensure that the full range of features has adequate representation. The row and column data in these datasets are in reference to the top left of the image with an origin coordinate of (1, 1). The challenge is to develop novel ways of processing these images to extract information that can help us monitor and manage coral reefs. Tabular dataset field descriptions: label_key.csv: id: label ID number label_code: the short code representing the features of interest functional_group: a more general categorisation of the features label_description: a brief description of what the labels represent benchmark: the classification accuracy (%) of the current best performing classification algorithm training_dataset.csv & validation_dataset.csv: qid: quadrat ID number row, col: the row and column of the pixel associated with the training/validation record (based on a top-left image origin) label: the names of the features of interest label_code: the short code representing the features of interest functional_group: a more general categorisation of the features filename: the name of the image associated with the training/validation record method: random (the point was generated randomly), target (the point was human- generated in order to ensure every feature has adequate representation Citation for this dataset: González-Rivero M, Beijbom O, Rodriguez-Ramirez A, Holtrop T, González-Marrero Y, Ganase A, Roelfsema C, Phinn S, Hoegh-Guldberg O (2016) Scaling up ecological measurements of coral reefs using semi-automated field image collection and analysis. Remote Sensing 8:30. License: These data are shared under a Creative Commons Attribution Share Alike License: https://creativecommons.org/licenses/by-sa/2.5/au/