This tensorflow.js demo removes people in realtime from video:
The web demo works great, albeit with a low FPS and latency. I’m assuming with some simple tricks, the opposite scenario should also be possible (single out people, remove background). I immediately could implement this in several of my projects with great effect, and also play with the uncanny boundaries of it, if given control of the right parameters.
How likely is it to see stuff like this in VDMX in the near future? I don’t know the technical frameworks well enough, so forgive me naive questions. And its not a direct feature request, I know you guys have enough to work on. But I’m curious, are there paths for tools like this to show up as effects/tools in VDMX?
Gave this some more thought, could this be programmed as an ISF filter (I assume it practically can’t), or then another format of plugin, is that technically feasible?
If the framerate/latency could be reduced enough for this to be used in camera performance (which I guess is the biggest challenge actually), I wouldn’t mind hiring someone to adapt this.
I don’t think this can be implemented as an ISF, though you can certainly do some similar ‘persistence’ ideas.
Short answer is that a lot of this stuff still runs not quite fast enough for real-time, especially if you have other things going on (but it is getting there and some stuff can). And that with designing VDMX we always have that balance of making something that does one thing very well vs exposing generalize capabilities to give people the ability to play around and customize.
Thank you, confirmed my suspicions and adjusted my expectations.