Automatic age estimation is a challenging problem attracting attention of the computer vision and pattern recognition communities due to its many practical applications. Artificial neural networks, such as CNNs are a popular tool for tackling this problem, and several datasets which can be used for training models are available.
Despite the fact that dogs are the most well studied species in animal science, and that ageing processes in dogs are in many aspects similar to those of humans, the problem of age estimation for dogs has so far been overlooked. In this paper we present the DogAge dataset and an associated challenge, hoping to spark the interest of the scientific community in the yet unexplored problem of automatic dog age estimation.

The DogAge Dataset

For sparking the interest of the scientific community in the challenge of dog age
estimation, we present and make publically available the DogAge dataset that
has been carefully collected in a collaboration between animal and computer
scientists. It contains images of dogs, mapping them to one of the three classes
young (0–2 years), adult (2–5 years) or senior (>6 years).

The dataset currently consists of two parts (see Table 1 and Figs. 1 and 2):

1. Expert data: contains 1373 images collected by animal scientists, sampling
   pet dogs, shelter dogs, laboratory dogs, working dogs and commercial kennel
   dogs. Their age and division into the three groups was carefully verified. The
   images are mostly high-quality portraits with the dog facing upwards forward.
   See Fig. 2 for examples.
2. Petfinder data: contains 26190 images collected using the APIs of Petfinder1,
   a portal for pet adoption. The division of dogs into groups is not verified, and
   there is a diversity of angles and distances of the photos. The raw data has
   been cleaned, removing photos with more than one dog, containing ther pets
   or large parts of humans, with resolution less then 500 px on any dimension,
   and signboard photos. See Fig. 3 for examples.

The two parts of the dataset together with the testing set can be found here.

Table 1. Image count by classes and datasets

		Adult	Senior	Young	Total
Expert data	Train/Val	370	495	223	1088
	Test	x	x	x	285
Petfinder	Train/Val	15083	2233	8874	15747

Citations

When using these datasets in your work, please cite our paper, Automatic Estimation of Dog Age: The DogAge Dataset and Challenge

@incollection{Zamansky2019,
  doi = {10.1007/978-3-030-30508-6_34},
  url = {https://doi.org/10.1007/978-3-030-30508-6_34},
  year = {2019},
  publisher = {Springer International Publishing},
  pages = {421--426},
  author = {Anna Zamansky and Aleksandr M. Sinitca and Dmitry I. Kaplun and Luisa M. L. Dutra and Robert J. Young},
  title = {Automatic Estimation of Dog Age: The {DogAge} Dataset and Challenge},
  booktitle = {Lecture Notes in Computer Science}
}