News for hackers. Stuff that matters.

A comment on HDR-NeRF: High Dynamic Range Neural Radiance Fields.

Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by ge96 on Thursday April 27, 2023 @01:27AM (#35723651)
I see these videos on YT, "walking through Tokyo 4K" stuff like that. It would be cool to have something that can watch the video and generate a street level map with depth of what it saw. You can then go through with a camera head level and navigate it. Can be done but yeah.
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by zexodus on Thursday April 27, 2023 @02:26AM (#35724036)
The technology would also be killer in real estate markets. Imagine being able to take a video and give potential tenants the ability to see the apartment/house/etc in detail.
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by ge96 on Thursday April 27, 2023 @04:03AM (#35724737)
Probably easiest solution is a 360 video camera since YT allows you to view them... but not sure how good of quality/distortions.
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by CyberDildonics on Thursday April 27, 2023 @09:31AM (#35727832)
You can already go on zillow and see 3d walk through by jumping from one spherical panorama to another. They even make 3d models out of it.
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by pornel on Thursday April 27, 2023 @10:28AM (#35728650)
I cynically think it wouldn't be a killer, because real estate agents prefer to have photos carefully made with fish-eye lens that makes rooms look bigger.
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by smusamashah on Friday April 28, 2023 @06:51AM (#35739663)
Check out zip nerf which seems to do exactly that https://jonbarron.info/zipnerf/
[ Parent | on HN ]
Re:HDR-NeRF: High Dynamic Range Neural Radiance Fields by ge96 on Friday April 28, 2023 @05:12PM (#35747170)
looks amazing, my naive take is ML is a brute force way to do something, not guaranteed to be correct but close enough

on the other hand, I'd be curious how much worse a procedural way to do frame by frame analysis would be

[ Parent | on HN ]