Speed
Pipeline even after including a number of steps runs in super real-time on consumer-grade CPUs .
Robust
Dedicated face-detector module makes it robust to large scale and resolution variations. Works on selfie videos and movie-collections alike.
End to End
Expects rgb[Uint8] data in a Numpy array and returns corresponding Embeddings along with corresponding bounding-boxes in absolute coordinates. Made for developers and creators.
Versatile
No versioning contraints, just Python3 and Numpy. Run it anywhere from a VPS to a local system.
Latency
Face-detection part of pipeline runs in constant time, but embedding would have to calculated separately for each face-detected leading to higher latency.
| Architecture | OS | Time*(ms) |
|---|---|---|
| Intel i5-8300H Cpu @ 2.30GHz | Windows | ~ 17ms |
* Time measured in python runtime averaging over 100 loops.
* Time measured with HD resolution as input with a single face present.