Terp

Terp (short for interpreter) is an image classifier tuned to identify the contents of an image. The image itself is unchanged, but the medtadata has the detections tagged on.

TODO

Per-frame classification, workgroup of N threads keeps it parallel. Passed through pipeline, stages are variable and determined by the number of classifiers available.
Go/No-Go on a per-clip basis for generalized detection of larger classifications
Motion is NOT detected at this stage. Motion detection is handled on an earlier stage of the pipeline.