Multimedia COMMONS

Community-Organized Multimodal Mining: Opportunities for Novel Solutions

A Workshop at ACM Multimedia 2015, Brisbane, Australia, October 26-30

Workshop Date: Friday, October 30, 2015

Update: Proceedings available now from the ACM!

Building flexible, accurate multimedia analysis and retrieval systems requires massive amounts of annotated data as ground-truth input. The MMCommons workshop will kick off the development of a research community around the Multimedia Commons Project (MMCP), an initiative initially focused on annotation of the 99.2 million images and nearly 800,000 videos in the Yahoo Flickr Creative Commons 100 Million dataset (YFCC100M) released in 2014. This freely available dataset is already being used as the basis for a public-domain corpus of computed audio, visual, and motion features (the YLI Corpus), a set of preliminary public-domain annotations for multimedia event detection (YLI-MED), and a visual search index.

The MMCommons workshop will provide a forum for the community of current and potential users -- i.e., everybody! -- to explore the possibilities for novel research offered by this unprecedentedly large, unprecedentedly open-source dataset; lay the groundwork for future data challenges and new benchmarking tasks that would not previously have been possible; and discuss priorities and plans for annotation and distribution based on community needs and interests. In addition to panels on research opportunities and possible data challenges, and demos and discussions of the planned annotation efforts, the workshop will offer an opportunity for researchers currently using the YFCC100M and the existing derived features and annotations to present papers about their work.

What the Human Genome Project accomplished for our understanding of human genetics, we anticipate the Multimedia Commons Project will do for consumer-produced images and videos: foster a fundamental understanding of the major underlying structures by amassing and analyzing detailed information about the building blocks. As we lay the groundwork for focused collaborations around an open, shared dataset, we hope to inspire participation from a diverse set of multimedia researchers working on a broad set of tasks. Major themes of the MMCommons workshop -- and the annotation efforts and data challenges coming out of it -- will include connecting information about large-scale and small-scale structure; new approaches to integrating modalities; comparing research that uses different datasets; interdisciplinary collaboration towards possible applications; leveraging high-performance computing; and innovative methods for generating subcorpora and bootstrapping annotations.

Looking for information about the Multimedia Commons Project in general? Check it out here!