Welcome

Welcome to MOONVID, a 3-year research project funded by the German Research Foundation (DFG). MOONVID is jointly conducted by the IUPR Research Group at the University of Kaiserslautern and the Multimedia Analysis and Data Mining Lab at the German Research Center for Artificial Intelligence (DFKI). This website provides information about the project and gives links to demos, press articles, and publications.
Feel free to look around!

Project Mission - Visual Learning from YouTube, Flickr, and co.

Currently, users world-wide generate and share larger amounts of images and videos than ever before in history, and huge web-based multimedia databases are evolving: YouTube sees another 20 hrs. of video every minute, Flickr 2,000 new pictures. This content is provided by a large community of users, who share it, comment it, debate it, label it, and categorize it.
With this information, YouTube, Flickr, and co. open new chances for the development of computer vision systems. While recognition systems have only been trained on small, manually labeled datasets so far, we can now perform visual learning on large datasets of web-based content. For example, to learn a concept (like "soccer"), material from YouTube and Flickr is acquired and analyzed using machine learning techniques. The resulting statistical detector can be applied to previously unseen images and videos - i.e., the system learns to detect soccer by watching YouTube. As this learning is completely unsupervised, more flexible and scalable recognition systems can be developed.

Show Case TubeTagger

lookapp

Smart Video Buddy

TubeFiler

Invire

Navidgator