Successful conclusion of the first FMVA Workshop at ICPR 2026

The first edition of the Foundation Models for Vision Applications (FMVA 2026) workshop was successfully held on 22 August 2026 in Lyon, France, in conjunction with the 28th International Conference on Pattern Recognition (ICPR 2026). The workshop focused on the growing role of foundation and vision-language models in computer vision, with particular attention to their evaluation and use in areas such as biometrics, multimodal learning, biomedical imaging, video analysis, and explainable AI.

The workshop was organized by the Department of Engineering of the University of Sassari, with Pietro Ruiu, Andrea Lagorio, and Seth Nixon serving as workshop organizers. The initiative brought together researchers working on different aspects of foundation models and their application to visual understanding, providing an opportunity to discuss current results, open issues, and emerging research directions.

The scientific programme featured seven oral presentations, covering topics ranging from multimodal large language models and audio-visual question answering to video action recognition, fair facial attribute classification, 3D reconstruction, egocentric visual grounding, and open-vocabulary object detection. The University of Sassari also contributed to the programme with the work “Are Pretrained VideoLLMs Enough for Action Recognition? A Zero-Shot Evaluation on Charades”.

A highlight of the workshop was the invited keynote by Prof. Arun Ross, one of the leading international researchers in biometrics and computer vision. His talk, “Foundation Models in Biometrics: From Matching to Explainability”, offered a broad and insightful view of how foundation models are changing biometric systems, moving beyond traditional matching tasks towards richer and more explainable forms of analysis. The keynote was particularly appreciated by the audience and provided an excellent reference point for the scientific discussions that followed.

The workshop attracted strong attendance throughout the morning and considerable interest from the ICPR community. The presentations stimulated active discussion and exchanges among speakers and participants, confirming the relevance of foundation models as a research topic that now cuts across many areas of computer vision and pattern recognition.

Following the positive response to this first edition, the organizers are considering a second edition of FMVA in 2027. The conference with which the next workshop may be held in conjunction has not yet been decided, but the aim is to continue building a scientific meeting point for researchers studying foundation models and their use in vision applications.