Logo

Siblings Database

Back to PROJECTS

SiblingsDB - Content description

The SiblingsDB contains different datasets depicting images of individuals realted by sibling relationships. Actually, the images are organized in two different DBs.

The first, called HQfaces, contains a set of high quality images depicting 184 individuals (92 pairs of siblings). A subset of 79 pairs contains profile images as well, and 56 of them have also smiling frontal and profile pictures. The images, with resolution 4256x2832, were shot by a professional photographer with uniform background and controlled lighting. The subjects are voluntary students and employees of the Politecnico di Torino and their siblings, in the age range between 13 and 50 (average 23.1). The highest and average age differences between siblings are 30 and 4.6 years, respectively. All subjects are Caucasian and 57% of them are male. Subjects were asked not to wear make-up. The images have been organized into three Individual Datasets (IDS), namely HQf, HQfp and HQfps:

All the images are annotated with, respectively, the position of 76 landmarks on frontal images and 12 landmarks on profile images. For each individual the information on sex, birth date, age and votes of the panel of human raters (who were asked to evaluate if the couples depict siblings or not) are also available. For each IDS, we created a Pair Dataset (PDS) for our experiments containing all the positive pairs, the N/2 pairs of siblings, and an equal number of randomly chosen negative, non-sibling, samples.

Examples of frontal, profile and smiling facial images of siblings (last row) and of non siblings (first row) in HQfaces

The second DB, called LQfaces, contains contains 98 pairs of siblings (196 individuals) found over the Internet, where most of the subjects are celebrities. The photographs have different resolutions (approximately from a few hundred to more than 3,000 pixels across). The poses are semi-frontal. Faces often show expressions (smile), and images were taken under different lighting conditions. Profiles are not available in LQfaces. The individuals are 45.5% male and mainly Caucasian. The position of the 76 frontal facial landmarks are provided as well, but this dataset does not include the age information and the HP ratings (human expert ratings were not collected since this dataset is composed mainly of well-known personages and, hence, likely to produce biased ratings). Data are organized in the same way as the previous datasets.

Examples of images in LQfaces

Experimental protocol

Two different experiments were performed in our work:

  1. within DB experiments. Each PDS (HQf, HQfp, HQfps and LQfaces) was classified assessing results using stratified five-fold cross-validation (CV); the accuracies reported in the paper are the average classification rates of the classifier over the different CV rounds.
  2. Cross-DB experiment. HQf PDS was used for training and LQfaces for testing

Please refer to the following paper for further details on the DB and on the experimental results:

T.F. Vieira, A. Bottino, A. Laurentini, M. De Simone, Detecting Siblings in Image Pairs, The Visual Computer, 2014, vol 30, issue 12, p. 1333-1345, doi: 10.1007/s00371-013-0884-3 Citation (BIB)

 

Download

SiblingsDB is available for free for non-commercial, research purposes. Please read the license agreement before proceeding and then download the DB from here. Please, let us know if there is any issue with the DB.

We recall that, according to the license, any published work that makes use of any part of the SiblingsDB must include a reference to the following publication:

T.F. Vieira, A. Bottino, A. Laurentini, M. De Simone, Detecting Siblings in Image Pairs, The Visual Computer, 2014, vol 30, issue 12, p. 1333-1345, doi: 10.1007/s00371-013-0884-3 Citation (BIB)

The complete package contains the following:

Related Papers