Official repository of the paper "Child Face Recognition at Scale: Synthetic Data Generation and Performance Benchmark".
In the paper we addressed the need for a large-scale database of children's faces by using generative adversarial networks (GANs) and face age progression (FAP) models to synthesize a realistic dataset referred to as HDA-SynChildFaces. The created HDA-SynChildFaces consists of 1,652 subjects and a total of 188,328 images, each subject being present at various ages and with many different intra-subject variations. An example is given below for a single subject:
The table below shows how the HDA-SynChildFaces compares to related datasets:
Contrary to other works, the HDA-SynChildFaces dataset is demographically balanced concerning several factors such as race and gender. To balance a demographic attribute, for instance, the subject's race, the subject is moved across the learned race boundaries as illustrated below.
The database is being made available for researchers from 2024 onwards. Interested researchers can download this database. Any commercial use/distribution of this database is prohibited. All the technical reports and papers that report experimental results from this database should provide acknowledgment and reference (see the Acknowledgement Section).
Link: https://cloud.h-da.de/s/pFpfkzbwkniS6gz
Password: bK7bBjp8myRs
The database folder contains 5 subfolders:
- Accepted: The original adult images post filtering (20+)
- Age group 0: Transformed images with age 16-13
- Age group 1: Transformed images with age 13-10
- Age group 2: Transformed images with age 10-7
- Age group 3: Transformed images with age 7-4
- Age group 4: Transformed images with age 4-1
Furthermore, each of the above is divided into a set of reference and probe images.
Demographic attributes is given as part of the filename, following the convention: <seed_age_sex_race> Where seed can be seen as a type of image ID which can be used to identify the subjects across age groups. For a description of the database generation, refer to the paper.
If you use this dataset or associated work, please cite the following paper:
@article{Falkenberg-HDASynChildFaces-Frontiers-2024,
title = {Child Face Recognition at Scale: Synthetic Data Generation and Performance Benchmark},
author = {M. Falkenberg and A. B. Ottsen and M. Ibsen and C. Rathgeb},
year = {2024},
journal = {Frontiers in Signal Processing}
}