@Lunayian The higher end PC grade models are a lot better at predicting where the wrists/elbows are when the view is blocked, but they use a lot of compute and it's a looot harder to block stuff from 2 cameras
@Lunayian Delay is already quite good, around half second to convergence for higher accuracy filtering and less so for high activity (cameras only 30hz btw!) - here's just running it on some random youtube video with a smartphone tier model, I need to do L/R flip correction & scary math :(
@Lunayian I control the environment, it's easier to add more data than it is to get accurate depth from 2 cameras very close or 1 camera like others are trying to do. I can just get 2d pose (AI models readily available and really good!!) and correlate the recognized keypoints in 3d space
@Ghost8099110002@ShinyQuagsire@zhuowei Probably in high cost-of-living areas -- Comes with high paychecks too, so the vision pro seems a lot less expensive compared to basic living expenses vs the rest of us
@MANAV_DHANJAL@Sashank24925305@GoPro The hero 8 is an excellent camera, the video quality difference won't be that great over the 8 though since the 9 is only doing 5k30 video and 4k60 still
@TechnicPack Any chance you could unlock one of my older accounts / change the email associated with it? I no longer own the domain associated with it and the last time I tried getting it fixed I got nowhere besides getting banned from the discord. https://t.co/4sFy5gAEhj