One of the most-requested features for the DeepBelief object recognition SDK has been integration with OpenCV. We’re actually heavy users of the framework ourselves at Jetpac, and so I’m pleased to say that it’s now easy to use it with DeepBelief!
Another frequent request was desktop support, and so there’s now Linux x86-64 and OS X libraries, documentation and examples available. I’m still a big fan of Caffe and Overfeat, but DeepBelief can be handy when you need simple setup with few dependencies, a very small footprint, and when you want to train models from file system images to use later on mobile devices.
It’s a very exciting time for computer vision, there’s a lot of new and exciting applications we can all build with the deep learning approach, and I hope this helps a few more people dive into creating their own!