KinSheep Dataset

KinSheep is a dataset for kinship detection from sheep faces.  It contains 8,642 high resolution jpeg images of 1,549 individual animals with pedigree covariance value between each pair.   

Data

Image data is split into 3 zipped folders data1.zip, data2.zip, and data3.zip, each containing the images and meta information.  Genetic distance between all pairs of animals is provided in relationships.csv. 

Image files

The format of the image file is _.jpg.  For example, 0003_002.jpg is the 2nd instance image of animal 0003.  Animal ids run from 0 to 1,548; the number of instances vary for each animal.

Metadata files

Each image is accompanied by xml information file _.xml.  For example, 0003_002.xml contains metadata for 0003_002.jpg.  Metadata provides information on the bounding box of the sheep's face in the image as well as the eyeline and noseline, which identify the location of the coarse features of the face.   

Face bounding box

The location of the face in the image is provided as the "sheepface" object with (xmin,ymin) and (xmax,ymax) coordinates of the top left and bottom right corners of the bounding box.  

Eyeline and noseline

The eyeline and noseline are features of the faces provided as the "eyeline" and "noseline" objects with (x1,y1) and (x2,y2) coordinates of the two ends of the line.  Eyeline goes from centre of one eye to the other crossing the bridge of the nose; if only one eye is visible, it goes from centre of that eye to bridge of the nose.  Noseline goes from the eyeline where it meets the bridge of the nose, along the bridge towards ending between the two nostrils; if the nostrils are not visible, it goes to the end of the nose.  

Relationships

The animals in the images are assigned (randomly generated) ids.  The file relationships.csv provide information about relationship between every pair of animals.  There are 4 columns: id1, id2, GenomicCovariance and PedigreeCovariance.  Ids identify the pair of animals for which the relationship is provided under genomic and pedigree covariance (pedigree relationship is an expectation and the genomic a realised estimate (which should be better)).  Pedigree of 1 means it's the same animal (anything over 1 indicates coefficient of inbreeding); around 0.5 indicates parent/child or sibling relationship, around 0.25 a grandparent/grandchild relationship, and so on.  

Citing KinSheep dataset

L. Szymanski and M. Lee. Coarse facial feature detection in sheep. In International Conference on Image and Vision Computing New Zealand (IVCNZ), pp. 1-6, 2021.

Acknowledgement

This project was made possible through the support of Beef+Lamb New Zealand.