This internship deals with the recognition (re-identification) of faces of children. ZiuZ is developing a face recognition system with the goal of identifying perpetrators and victims in illegal images and videos. In academia and industry, state-of-the art face recognition systems are based on deep neural networks. These models are largely trained on adult faces, making the application to faces of children more difficult.

THE ASSIGNMENT

Your assignment will focus on analyzing and improving the performance of a face recognition system and in particular enabling it to better recognize children’s faces. During this internship, you will: 

• Compile a dataset of faces of children, using existing datasets as a starting point.
• Create a benchmarking tool/script to analyze the performance of face recognition models used at ZiuZ on this dataset.
• Evaluate existing candidate models/algorithms tailored to child face recognition.
• [Optional] Try to improve (e.g. by finetuning) face recognition models using faces of children.

EXPECTED END RESULT

1. Benchmarking tool (evaluating models on relevant data)
2. Report on accuracy of current pipeline on above benchmark(s)
3. Literature review on child face recognition
4. Improvement on current models or implementation of a novel system for faces of children.
5. Final report combining (2) and (3) with documentation of step (4).

TASK DESCRIPTION / INTERNSHIP REQUIREMENTS

What is expected of you?

• Self-initiative
• Familiarity with
  • Computer vision
  • Machine earning and deep learning
  • Deep learning frameworks (PyTorch, Tensorflow)
• Proficiency in python (or C# or C++)

REMUNERATION
ZiuZ has a standard internship / graduation allowance of € 275 per month and a maximum travel allowance of € 200 for travel by car. Public transport costs are reimbursed in full.