If one is searching for a solution to do image recognition, the usual development stack will be a computer that runs open-cv, and a web camera. There detection mechanism is purely based on the predefined set of features, most of the time geometry, or distances between the point clouds of interest. The were some interesting project that came with such image recognition stack, such as this https://www.instructables.com/id/Face-tracking-gun...

The main gripe with such stack is that the recognition is solely based on the existing date loaded in to it. A classic issue faced: what happened if the subject change state over time, will it still be recognize. Example: picture of a dog loaded in the stack, and picture of a wet dog (the same dog) acquired from the web cam; picture of human loaded in the stack, and picture of older human (the same human) captured from the web cam. Humans will instinctively recognize the said subject, but computers need to be trained and predict the possibility.

DeepLens come to the rescue!!! It is essentially an Intel Atom based computer that connects to the Internet via WiFi, runs image recognition with Deep Learning while connecting to AWS suite of software layers. Intuitively It is somewhat similar to the stack mentioned above, but now on steroids. The classic difference between DeepLens vs any other stack, is the ability to capitalize on the infinite computing power of a cloud (to train models that is for the choice of Machine Learning algorithm). DeepLens is beyond that, it is a powerful edge computing device targeting specifically at machine vision enabled computing devices.Think of the many uses cases make possible with DeepLens!!

In the deep lens workshop at re:invent 2017, the participants were given an opportunity to come close and hands-on with

Please note: this is a work-in-progress post