Deep-Learning for Gender Classification/Age Estimation

Status: available
Supervisor: Martin Kampel

Problem Statement

Automatically detecting the gender of a person or estimating his/her age is valuable in many fields of work.
In the area of computer vision exist a variety of possible approaches, mostly based on machine learning, to accomplish this task.
A higher complexity arises due to the claim to classify in-the-wild images, i.e. pictures which do not fulfill certain restrictions, usually found under laboratory conditions.
Not fulfilled conditions for in-the-wild images could be (without any claim to completeness):

  • 0-n faces visible
  • partially occluded faces
  • faces in non-front view
  • etc.


The goal of this work is to train and evaluate a given convolutional neural network [1] to accomplish gender classification and age estimation on images found in the wild (e.g. web images). As a starting point, the CNN proposed by Hassner et al. [1] should be trained and evaluated.
This comprises certain tasks as finding and processing a suitable dataset for training, evaluating and testing the network, as well as tuning the hyperparameters, and evaluating the solution.


  • Literature Review – CCNs in general, Levi/Hassner, VGG-16, or other feasible solutions
  • Dataset research and processing
  • Preprocessing of the images
  • Recurrent Training and Evaluation
  • Final Evaluation
  • Written Report/Thesis and final presentation


  • Experience with python (NumPy, torch7, …)
  • Goal-oriented working method
  • Machine learning knowledge
  • Scientific writing
  • Deep learning knowledge beneficial (CNN, keras, theano, tensorflow, etc.)


[1] Levi, Gil, and Tal Hassner. “Age and gender classification using convolutional neural networks.” Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops. 2015.