"Our landmark recognition engine, in fact, can provide input data to these 3D reconstruction systems and enables them to be scalable to a large number of landmarks."

This might be revolutionary, as being one of the first automated systems to aggregate visual data into intelligent sets, making it scalable from local to global.

http://www.google.com/googleblogs/pdfs/google_landmark_recog...

How does this compare technically with Microsoft's Photosynth project? http://photosynth.net/ ... looks like they're already doing what Google proposes to do.

Photosynth aggregates images already known to be of a particular scene and figures out how they fit together. This paper is about taking an image from an unknown place and identifying whether it is of a particular landmark.

Both probably make heavy use of unsupervised clustering techniques on images but are otherwise totally different use cases, if nothing else.

"Google: we brute-force stuff."

The fact that this is basically true makes it all the funnier!

I was super impressed in the street view - navigation updates where the circle changes on depth, and goes sideways on buildings. Really how long until they can upgrade the lasers to work good enough to scan a building front (barcode style).. 5 years? Until then this a great hack-ish way of doing it