Deep Expectation of Real and Apparent Age from a Single Image Without Facial Landmarks

In this paper we propose a deep learning solution to age estimation from a single face image without the use of facial landmarks and introduce the IMDB-WIKI dataset, the largest public dataset of face images with age and gender labels

Rasmus Rothe

2016

Scholarcy highlights

  • Age estimation from a single face image is an important task in human and computer vision which has many applications such as in forensics or social media
  • In this paper we propose a deep learning solution to age estimation from a single face image without the use of facial landmarks and introduce the IMDB-WIKI dataset, the largest public dataset of face images with age and gender labels
  • We validate our methods on standard benchmarks and achieve state-ofthe-art results for both real and apparent age estimation
  • The ChaLearn Looking At People International Conference on Computer Vision 2015 challenge provided the largest dataset known to date of images with apparent age annotations, here called the Looking at People dataset, and 115 registered teams proposed novel solutions to the problem
  • For all experiments we report the Mean Absolute Error in years
  • For reference in Tab. 1 we report the performance when employing standard Support Vector Regression with RBF kernel and -insensitive loss function on deep features extracted from the last pooling layer, last and penultimate fully connected layer of our deep architecture without and with pre-training on IMDB-WIKI dataset
  • Pre-training on the IMDB-WIKI dataset results in a large boost in performance suggesting that the lack of a larger dataset for age estimation was overdue for a long time

Need more features? Save interactive summary cards to your Scholarcy Library.