Idiap Institute releases report on applying foundation models in biometrics

Over the past few years, the world has been witnessing the power of foundation models, machine learning models trained on immense datasets that complete various general tasks, including conversing with humans such as in the case of ChatGPT. But what are foundation models bringing to biometrics systems?

A new report from the Switzerland-based Idiap Research Institute provides an overview of current research on using foundation models in biometrics. The paper, titled Foundation Models and Biometrics: A Survey and Outlook, also outlines advancements that can be introduced to biometrics-related problems such as recognition accuracy and enhancing security and privacy.

“To our knowledge, this paper is the first survey which investigates the applications of foundation models in biometrics,” Sébastien Marcel, senior researcher at the Idiap Research Institute, announced on LinkedIn on Thursday.

Aside from Marcel, the paper was co-authored by Hatef Otroshi, a postdoctoral researcher at the institute.

Foundation models are currently being tested in different domains, including natural language processing, computer vision, audio processing and multi-modal processing. Some studies, for instance, have shown that pre-trained multimodal large language models such as GPT-4 can recognize face and iris data, without any additional fine-tuning.

Several benchmarks have also demonstrated that pre-trained foundation models can be effective for biometric tasks while some papers have attempted to fine-tune pre-trained models for biometric problems. Other researchers have used self-supervised learning to learn biometric representation from unlabeled data and evaluated the models for downstream tasks.

The research, however, also highlights the limitations in applying foundation models in biometrics.

“We can envision more future work on developing biometric foundation models with broader applications and better performance for different biometric tasks,” the paper notes.

The Idiap Research Institute is organizing the first workshop exploring the foundation models for face and gesture recognition tasks, under the name International Workshop on Foundation, Multimodal Large Language and Generative Models for Face and Gesture Recognition (FM&LLM&GM2025). The workshop will be held during the IEEE conference series on Automatic Face and Gesture Recognition between May 27th and 29th.

Aside from the workshop, the topic will also be discussed at the Martigny Biometrics Workshop between May 19th and 20th, organized by the institute alongside the European Association for Biometrics (EAB) and the Center for Identification Technology Research (CITeR).

Article Topics

biometrics  |  biometrics research  |  foundation models  |  Idiap  |  machine learning  |  research and development  |  Swiss Center for Biometrics Research and Testing