Hacker Newsnew | past | comments | ask | show | jobs | submitlogin

It uses mediapipes selfie segmentation module: https://google.github.io/mediapipe/solutions/selfie_segmenta... using MobileNet v3 (around 0.5 GFlops of juice it's gurgling, so enough to run on a raspberry, maybe not your Casio)


It'd be interesting if this could be embedded in the client/server, e.g. Jitsi. At work we turn off cameras to save bandwidth, but this seems like you could just send a bitstring, and heavily compressed at that.


Actually that reminds me of - and i personally think it's the future of video compression - that there a neural network based codecs in the making https://developer.nvidia.com/ai-video-compression


Yeah would be fun if you could gzip/compress let's say 4 frames every 200ms and make your own ascii video stream codec so to speak




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: