Background/Motivation:
Face-based age estimation is central to many applications (e.g., identity verification, youth protection, medicine). Classical approaches (pure regression or simple classification) have clear limitations, however: they ignore uncertainty, suffer from imbalanced data (long tail, missing age classes), and the non-linear scale of age. At the same time, a recently published study claims that the choice of loss function and architecture only has a limited impact on performance [7]. This blanket statement is to be critically and thoroughly examined in this work to clarify when the choice of loss and architecture is decision-relevant and when it remains secondary.
Objective: The aim of this master's thesis is a systematic comparison of key modelling approaches and loss functions for age estimation, as well as the development of new, robust methods.
To this end:

• Approaches to be implemented and compared: classical point regression, probabilistic regression, and quantile regression [4], classification [1], ordinal classification [2,3], label distribution learning [5], hybrid methods, etc.
• Conditions and problematic cases to be examined: missing or small age classes, strong imbalance (long distribution tails), label noise, non-linear age scale.
• Existing metrics are critically examined and suitable metrics are identified or developed.
• New procedures will be designed and evaluated based on the new insights.

Results: Robust guidelines are expected on when which approach works (or fails), including ablation studies on loss design, binning/ordinalization, distribution targets, and hybridisations. The work provides reproducible implementations, strong baselines, extensive evaluation on common datasets (e.g., UTKFace, IMDB-WIKI, APPA-REAL, MORPH II, AgeDB), as well as proposals for new, more robust methods.

The work presents robust guidelines on when which approach works or does not work, limitations and pitfalls, and unexpected results. The methods are evaluated and compared using publicly available benchmark datasets and self-developed scenarios. The code used is well-documented, reusable, and the results are reproducible.

Be part of change

• Researching and structuring the literature on a current topic in the field of machine learning.
• Researching and implementing novel machine learning and computer vision approaches.
• Planning and conducting experiments.
• Self-critical evaluation of the obtained results.
• Presenting the results.
• Preparing a scientific paper in the form of a master's thesis with the results. 

What you contribute

• Good knowledge in the field of machine learning and training neural networks.
• Ideally, knowledge in computer vision and facial recognition.
• Good Python skills, preferably experience with PyTorch, OpenCV, etc.
• Motivation to independently delve into new and current research topics.
• Interest in optimisation and evaluation metrics.
• Interest in scientific research. 

What we offer

• Independent work schedule management
• Insights into the intersection of academic research and industrial application 

 Related works: 
[1] Roth et al. (2015). Deep EXpectation of Apparent Age from a Single Image  
[2] Cao et al. (2019). Rank consistent ordinal regression for neural networks with application to age estimation
[3] Niu et al. (2016). Ordinal Regression with Multiple Output CNN. CVPR.
[4] Shen et al. (2024) Nonparametric Estimation of Non-Crossing Quantile Regression Process with Deep ReQU Neural Networks. JMLR.
[5] Geng et al. (2013). Facial Age Estimation by Learning from Label Distributions. TPAMI.
[6] Kuprashevich et al. (2023). MiVOLO: Multi-Input Transformer for Age and Gender Estimation.
[7] Paplhám et al. (2024) . A Call to Reflect on Evaluation Practices for Age Estimation: Comparative Analysis of the State-of-the-Art and a Unified Benchmark. CVPR.

We value and promote the diversity of our employees' skills and therefore welcome all applications – regardless of age, gender, nationality, ethnic and social origin, religion, ideology, disability, sexual orientation and identity. Severely disabled persons are given preference in the event of equal suitability. Our tasks are diverse and adaptable – for applicants with disabilities, we work together to find solutions that best promote their abilities. 

With its focus on developing key technologies that are vital for the future and enabling the commercial utilization of this work by business and industry, Fraunhofer plays a central role in the innovation process. As a pioneer and catalyst for groundbreaking developments and scientific excellence, Fraunhofer helps shape society now and in the future. 

Ready for a change? Then apply now and make a difference! Once we have received your online application, you will receive an automatic confirmation of receipt. We will then get back to you as soon as possible and let you know what happens next.

Fraunhofer Institute for Secure Information Technology SIT

Requisition Number: 82686                Application Deadline: