TThe Video Browser Showdown (VBS) is an annual international competition for evaluating interactive video retrieval systems in a live, comparative setting. Held as a special event at the International Conference on MultiMedia Modeling (MMM) since 2012, VBS brings together research teams that develop advanced video search tools and test them under the same conditions: the same dataset, the same tasks, the same time limits, and the same evaluation infrastructure. Its goal is to push research on large-scale, content-based video retrieval systems that are effective, fast, flexible, and easy to use.
VBS focuses on realistic search scenarios in large video archives, including Known-Item Search (KIS), where a specific video segment must be found; Ad-hoc Video Search (AVS), where many relevant examples for a general topic must be retrieved; and Visual Question Answering (VQA), where participants must answer content-related questions based on the video collection. The competition includes both expert sessions, performed by the system developers, and novice sessions, performed by users from the conference audience, thereby evaluating not only retrieval performance but also usability.
Over the years, the VBS datasets have grown substantially, from tens of hours of video in the early editions to thousands of hours in recent competitions. Recent editions use large-scale and diverse collections such as the Vimeo Creative Commons Collection (V3C), the Marine Video Kit (MVK), and medical video data from laparoscopic hysterectomy procedures (LHE/VBSLHE- i.e. GynSurg). VBS 2026 used nearly 4,000 hours of video content from V3C, MVK, and LHE datasets, demonstrating the increasing scale and diversity of interactive video retrieval evaluation.
Each task is time-limited and scored based on factors such as search speed, correctness, penalties for false submissions, and – in AVS tasks – the diversity and number of relevant results found. By combining a live competition format with standardized logging and evaluation via DRES, VBS provides a unique benchmark for comparing interactive video search systems and for identifying trends in multimedia retrieval research.
VBS Events
| Year | Location | Number of teams |
|---|---|---|
| 2026 | Prague, Czech Republic | 17 |
| 2025 | Nara, Japan | 17 |
| 2024 | Amsterdam, The Netherlands | 12 |
| 2023 | Bergen, Norway | 13 |
| 2022 | Phu Quoc, Vietnam | 16 |
| 2021 | Prague, Czech Republic – virtual | 17 |
| 2020 | Daejeon, Korea | 10 |
| 2019 | Thessaloniki, Greece | 6 |
| 2018 | Bangkok, Thailand | 9 |
| 2017 | Reykjavik, Iceland | 6 |
| 2016 | Miami, USA | 10 |
| 2015 | Sydney, Australia | 8 |
| 2014 | Dublin, Ireland | 6 |
| 2013 | Huangshan, China | 6 |
| 2012 | Klagenfurt, Austria | 7 |
All information about the Video Browser Showdown can be found here.
diveXplore – Distributed Interactive Video Retrieval System
diveXplore is our system that has been developed for interactive video retrieval by Klaus Schoeffmann, Mario Leopold, and Sahar Nasirihaghighi. It was also the best performing retrieval system for the question-answering session at VBS 2024.
