TLD2.2 is optimized on both the algorithmic and hardware level. We have reduced significantly the footprint of our detector which resulted in more efficient use of cash. We are also using SSE4 and NEON instructions to push the performance to limits on current hardware.
The resulting speed of TLD2.2 depends on the resolution of the input video, tracking profile, number of targets tracked in parallel and the target platform. The table below shows the results of the benchmark we have performed: