3 ms·
According to [1], the key differentiator is parallax, i.e. the positions of objects change relative to each other, depending on how the viewer is looking at the
by abrichr 3y ago
According to [1], the key differentiator is parallax, i.e. the positions of objects change relative to each other, depending on how the viewer is looking at them.
[1] https://www.youtube.com/watch?v=pmqr6ydIzDA https://www.youtube.com/watch?v=pmqr6ydIzDA
- two_handfuls 3y agoSee that information would have been super helful in the article. Thank you!
- avalys 3y agoHow is that achieved by capture on the iPhone 15 with the two lenses just 2 cm apart?
- abrichr 3y agoMy guess is Neural Radiance Fields [1], or something quite similar. https://www.matthewtancik.com/nerf https://www.matthewtancik.com/nerf
- dTal 3y agoDefinitely not anything like that. You need dozens of calibrated images, long training time, and - until Gaussian Splats - non-realtime reconstruction. This is something relatively trivial that runs in realtime on nothing more than stereo pairs. There's simply not enough information to extract anything more than a crude depth map.
- dTal 3y agoI would imagine by calculating a depth map from the stereo pair, then reprojecting onto that. This would give the "Minority Report" look described. No mention of artifacts in the article, but unless some clever AI infilling were done it would be very obvious with even small head movements.
- two_handfuls 3y agoThe next question is: how far can we move before the image shows obvious artefacts?