I am very grateful to the SeabirdWatch project, who have ownership of all data used in this project, and have kindly let me use it. As such I have no ownship over this data, nor any right to publish/share it. Instead, I provide a log here of the data used in this project.
All images in the dataset were taken from a single camera, located on Hvitabjarnarey Island in Iceland (colony code 'HVITa'). The images depict a relatively small colony, with approximately 30 nests per photo.
The full set of original images was provided, each with an approximate size of 500kb per (1920 x 1080) image. These images were compressed for upload to the citizen science programme Zooniverse, to an approximate size of 100kb per (1000 x 562) image. As all annotations have been completed on this dataset, I will train and test on this dataset.
The dataset used is a combination of data collected in 2016 and 2018, randomly separated into testing and training datasets.
I use collated annotation data from the citizen science platform, and obtained following the methods used in Jones et al. 2018. In summary, at least ten users will classify every image, whose raw clicks are combined to give one ‘consensus’ click per bird, in the form of x and y pixel values in the originial size image. The number of raw clicks that each consensus click is formed from is recorded for each image, but currently I do not weight each data point according to this.