We are excited to announce the availability of a visual search index on top of the Yahoo! 100M CC-licensed Flickr images corpus. This search index enables the fast execution of visual similarity search queries, i.e. given an arbitrary input image, it will return the most visually similar images out of the 100M images of the corpus.
Processing and indexing such a large amount of images was possible thanks to the high-quality feature extraction and indexing implementation developed by SocialSensor, and their deployment on top of a large cluster of computing nodes using Elastic Map-Reduce. But querying the index can be done locally on a commodity server (16GB of RAM is sufficient) once you download this index!
We believe that giving the possibility to everyone to perform fast visual similarity search over such a massive collection of images could be used in a number of core multimedia analysis and retrieval tasks, e.g. multimedia annotation, landmark recognition, geolocation estimation, etc.
More details on downloading and using the dataset are available here.