For an example of this in action check out the TV Comment Bot project.
SyphonDeepBeliever is a simple Mac utility app for connecting video feeds to the DeepBelief neural network SDK and outputting the results over OSC. Video feeds can come from any Syphon enabled application or from a live camera input. The results are provided for output both as a set of ordered label strings and values for the top ranked predictions and as raw levels for each of the detected strings.
To use SyphonDeepBeliever, choose a Syphon or live camera input from the “Sources” list in the main interface. The “Threshold” level can also be set to adjust the number of resulting predictions.
The Preferences panel contains the options for setting the OSC destination and other settings. Bonjour enabled OSC applications can be auto-setup from the “Detected Destinations” menu, or the IP / Port can be manually entered. An additional base OSC address can be set as a prepend for all outgoing messages for routing purposes.
This software is provided completely free, as-is, with no warranty or guarantees.
SyphonDeepBeliever requires Mac OS X 10.9.5 or later
Download SyphonDeepBeliever 0.1 for Mac OS X (released Jan 25, 2015)
Many thanks to MrRay for VVBufferPool and VVOSC libraries, Vade + Bangnoise for making Syphon, Pete Warden for DeepBeliefSDK and Apple for AVFoundation. This project basically smashes together examples from these wonderful codebases.