# How far ahead of academia is industrial image recognition? I don't work at [Google, Microsoft or Facebook] but I would guess: not that much ahead but much more polished. For instance, the top team on ImageNet at ECCV was SuperVision. The deep learning they were using is similar to what Google does, and Google actually hired the team. By the way, Google's CV research is not really what I would call "undisclosed", they do publish interesting papers. See for instance [this one](http://static.googleusercontent.com/media/research.google.com/en//pubs/archive/40814.pdf). An example of something not so public, and hard to reproduce for obvious reasons, is what they are doing with their D-Wave quantum computers... Personally I have been working at a small startup CV for about four years. We used to do image recognition on the server "à la Google Goggles", these days we focus on image recognition on mobile devices. In terms of algorithms, we often do some things slightly different than the state of the art papers and Open Source libraries (e.g. OpenCV) because we have different goals. But what *really* differs is the quality and performance of the implementation. That, and we ended up being quite good at tuning parameters, finding tricks to work around problems... basically making trade offs.