MOONVID Project Info

The overall goal of MOONVID is to investigate how the statistical analysis of web-based multimedia content can help to improve retrieval systems. On this page, you find information related to the research goals, potential applications, evaluation efforts, etc.

Research Focus

In MOONVID, we particularly focus on aspects related to web-based video content:

  1. What are appropriate visual features and machine learning techniques to model large-scale diverse video datasets such as YouTube content?
  2. User-generated annotations are less reliable than explicit manual annotations. How can visual learning be robust with respect to incomplete, error-prone, and inaccurate training data?
  3. Users do not only tag their videos, but also categorize and rate them. How can this information be used to improve recognition systems?
  4. How can motion information contained in web video contribute to an improved recognition?


For a list of publications, please see the "publications" section.


MOONVID is a research-centered project, but applications of our technology can be found in various areas, like image and video retrieval, multimedia asset management, digital forensics, recommendation, and many others. For selected demonstrators of our technology, please see the "demo" section.

Image Understanding & Pattern Recognition Group
Multimedia Analysis and Data Mining Group