Retrieve and Segment: Are a Few Examples Enough to Bridge the Supervision Gap in Open-Vocabulary Segmentation?
Bridging the supervision gap in open-vocabulary segmentation using a few pixel-level annotated examples along with class names.
PhD Student at CTU in Prague (VRG). Instance-level recognition, open-world segmentation and retrieval.
Retrieve and Segment received the Best Oral Presentation award at the Greeks in AI Symposium 2026 (Athens, 15–17 July 2026). I also volunteered at the event for the second year in a row.
Greeks in AIPresented Retrieve and Segment as a Highlight paper at CVPR 2026 (June 2026).
Browse the Project pageOfficial implementations of Retrieve and Segment (CVPR 2026) and Only-Style (ICCV-W 2025) are on GitHub.
GitHubAttended the ELLIS Winter School on Foundation Models in Amsterdam, 24–27 March 2026.
Event detailsOur work Only-Style won the Best Student Paper Award at the Personalization in Generative AI (P13N) Workshop, ICCV 2025.
Read the paperI am a second-year PhD student in the Visual Recognition Group (VRG) at CTU in Prague, under the supervision of Giorgos Tolias, working on instance-level recognition, open-world segmentation and retrieval. In a few words, I am trying to create models that recognize (retrieve, classify, detect, segment) particular objects in an open-world environment.
Bridging the supervision gap in open-vocabulary segmentation using a few pixel-level annotated examples along with class names.