This is a great writeup. We sent our Chief Scientist to CVPR and he came away with a lot of the same conclusions - and now he wants to apply CNN to teach our cloud-based large scale SLAM map better real world objects :P.

More broadly though it was clear that most of the really good real time environmental applications for CV (SLAM etc...) are being applied to autonomous navigation. That's great but it crowds out investments in the stuff like we (Visidraft) are doing with AR and makes hiring and finding investors harder.

Damn. I thought that was my idea. Slam is ripe for some deep learning love. So many ways to apply it. God if only I didn't have a job. I've considered going to prison for a minor and morally acceptable offence if only so I could have time to work in this field - and I would but they don't have computers in prison. Perhaps a permissive monastery will take me in. Perhaps in another partition of reality, somewhere out there, is a version of me that has the time to play in these glistening pools. Such a shame to miss the start of something even bigger than the internet...