Now that we have real-time face recognition, one major bottleneck is the interface between the python script and the face recognition system -- the cropped images must be saved to disk and then read by face-rec. If we use the C++ API for OpenCV we will have much finer control over the pipeline, and the data will reside only in memory. Between this and using the GPU for inference, we may be able to solve our performance problems.
Now that we have real-time face recognition, one major bottleneck is the interface between the python script and the face recognition system -- the cropped images must be saved to disk and then read by face-rec. If we use the C++ API for OpenCV we will have much finer control over the pipeline, and the data will reside only in memory. Between this and using the GPU for inference, we may be able to solve our performance problems.