Australia Animal Species Image Dataset

Annotated images of 47 Australian animal species with taxonomic classification

Source
Kaggle
Country
Australia
Region
Oceania
Format
multiple formats
Access
Account needed
Themes
Birds & Wildlife, Biodiversity & Ecosystems, Machine Learning & AI

Get the data from Kaggle

About this dataset

This dataset contains 47 different Australian animal species. Every category is names as the serial number plus animal’s scientific name which based on taxonomic ranks. All these species are observed in Australia, some of them are Australian endemic species.

Every category folder includes three subfolders: annotations, images with backgrounds, and images without backgrounds. Each of these subdirectories comprises 600 files.

The ’images with background’ subdirectory contains 600 original images of the animals, complete with their respective backgrounds. The original meta-images are taken from the Global Biodiversity Information Facility (GBIF) and have been manually screened to remove images that do not meet the requirements, such as highly occluded images, animal corpses, scratches, specimens, excrement, etc.

The ’images without background’ subdirectory contains images from which the backgrounds have been removed, leaving only the original animal subjects.

Lastly, the ’annotation’ subdirectory consists of 600 files in JSON format, each encompassing detailed information about the original image, such as the species’ scientific name, label identification, and metadata concerning the animal’s spatial orientation within the image.

It is worth noting that the meta images of some species are less than 600 images due to various factors, such as individual size, population size, habitat and region, etc. Therefor, we applied image augmentation (horizontal flip and vertical flip) for those species.

We use Efficient Interactive Segmentation (EISeg) to assist in the manual labelling. Each original image will be saved as four files after manual labelling, including the image with the background removed, a polygon of the label, a JavaScript Object Notation (JSON) format file and a pseudo-colour image.

Browse sustainability datasets on SusanHub