The maximum number of streams and number of faces depends on
- The power of your CPU and whether it supports the AVX2 instruction set.
- The power of your NVidia GPU, typically measured by the number of CUDA Cores.
- The resolution of the video streams you wish to analyse.
As a quick guide though, take the following hardware
- Intel E3-1515M v5 @ 2.80 GHz. This has a CPU Benchmark of 10093.
- GeForce GTX 1050 Ti. This has 768 CUDA Cores and 4Gb VRAM.
- 8 GByte Ram.
- 256 SSD Hard Disk.
- Running Linux Identifier.
This hardware is powerful to support the analysis of
- Two Full HD Streams (1900x1080). This will analyse video at the required rate of 8 frames-per-second.
- Up-to 10 faces per stream (i.e. 20 in total across both streams). When a full 20 faces are in the video scene we would expect to utilise about 90% of CPU power and 70% of GPU power.
- Identifier will still work on more faces-per-stream, however the frame-rate of the video processed will drop.
- To analyse multiple streams, you a licence that supports it. The demo licence provided with Identifier will analyse upto 2 concurrent streams.
- These figures releate to using FaceDetector2 as the main face-detector engine. In Identifier 1.1 this is the default face-detector.