Open alexandre-leites opened 3 months ago
Maybe a quick win could be, not to scan the whole scene, and only have the detection on a specific part activated. Like shown in the following picture, that should also increase the performance
im implementing this now, shuold be out tonight(-ish)
implemented in #17
Would be nice to have the option to scale the frame being analysed to a smaller resolution, send it to the pipeline, and scale back the results.
In certains case, the scaling is way less resource intensive than analysing a full frame image/video. For example, you could scale a 4K video to 720p, send it to analyses, and scale back the "coordinates" of the mask before aplying it, so you can have the mask work on the original source without wasting resources analysing a full 4K frame.
I have seen this on other projects and it normally works fine depending on what you are trying to track or find in the frame.