Field of Research
We believe vision systems should be as dynamic and adaptable as the world they perceive, a world in constant change. In the SEEALL group, we advance computer vision research by building systems that learn from limited data, adapt swiftly to new situations, and uncover meaningful insights. We strive to make visual intelligence accessible to everyone tackling unique challenges, no matter the resources.
Scarce-data Computer Vision
Modern computer vision models depend on large, meticulously labeled datasets, yet many specialized fields, like medical- or scientific imaging, lack such resources as annotation costs are high due to the limited availability of experts. Our research aims to enable robust visual understanding with minimal annotation effort. We investigate learning paradigms that effectively utilize heterogeneous, low-cost supervision, such as semi-, weakly- or self-supervised learning, to bring vision systems to flourish even in scarce environments.
Adaptability in Vision Systems
Modern vision systems are often designed for static, predefined tasks, yet the real world demands flexibility. When deployed, these models encounter distribution shifts, new task variations, and evolving user needs, requiring costly retraining or complete model redesign. Our research focuses on building adaptable vision systems that can generalize to new vision tasks and user guidance on-the-fly. To do this, we explore visual in-context learning and study how these vision models perform in unusual, out-of-domain conditions.
Enable Vision in Niche Domains
Many specialized domains, from medical imaging to scientific research, need vision systems tailored to their specific applications. In our research we develop vision systems with niche applications in mind and work towards models that adapt to one-of-a-kind problems without extensive retraining or resources. We explore how generalist models can be customized on-the-fly through visual in-context learning and efficient test-time training, making visual intelligence accessible across diverse applications, allowing for specialized solutions to be built with minimal effort.
Find further information on our research page.
News
Pradnya Halady joins!
June 3, 2026
We warmly welcome Pradnya Halady, who joins as the first doctoral researcher in the SEEALL group!
In her research she will focus on visual in-context learning and test-time adaptation for large vision and vision-language models.
Her research will enable models to efficiently adapt to new, domain-specific visual tasks using a small number of contextual examples while balancing performance and the compute used.
Read more
Group Kick-Off
April 1, 2026
Welcome to the website of the Sample-Efficient and Emergent Adaptative Lightweight Learning (SEEALL) group.
April 1st, 2026 marks the kick-off of this KiKIT-funded young investigator group led by Simon Reiß that aims to push the frontiers of data-efficicent and adaptable vision systems.
Read more