There have been a lot of crazy many-camera rigs created for the purpose of capturing full spatial video. I recall a conversation at Meta that was basically “we are going to lean in as hard as possible on classic geometric computer vision before looking at machine learning
@ID_AA_Carmack How do you feel about the term "special"? I still find myself rejecting it outright, due to it feeling like a forced term when Apple refuses to use the VR/AR/MR terms and made up their own nonsense instead.
@ID_AA_Carmack Why not steal from radars and spin the camera at high speeds with a "perfect lens set up" and then use software to stich the image and remove any preceived movment blur?
@ID_AA_Carmack @jimmy6DOF The light field photography was very promising but data streams were too fat.
@ID_AA_Carmack Fascinating to see Meta's pragmatic approach to spatial video. Starting with geometric CV before ML mirrors the $BTC ethos – proven fundamentals first, then layer on complexity. Sometimes the old ways ARE the right ways. 🎯
@ID_AA_Carmack John when will it be feasible to put something like this in orbit, maybe adjacent to the iss, so nerds like me can experience orbit in VR?
@ID_AA_Carmack "and of course you want to maximize your use of the ground truth!" is a good quote.
@ID_AA_Carmack Omni-video + NeRFs is the future, the more data the NeRF has the better the 3d recreation.
@ID_AA_Carmack John, I want to see you on Lex Fridman, Rogan or another long form podcast discussing state of the art tech soon! There are two guys who constantly blow my mind on podcasts - you and Marc Andreesen.
@ID_AA_Carmack Would you agree that Meta have the most advacned volumetric capture rig on the market? I doubt there is anything else quite like what they developed for Codec Avatar. Their Emergent camera system is quite something.
@ID_AA_Carmack I know I'm a nerd because I'm just thinking of how to optimize the storage throughput from all that sweet, sweet data.
@ID_AA_Carmack That’s quite the camera rig! Agree that immersive is the way. Just feels so natural.
@ID_AA_Carmack yeah wouldn't surprise me
@ID_AA_Carmack How does the brain wire an 8 eyed spider 🎥
@ID_AA_Carmack It's sort of fun that every few years I get tapped to do various tests with immersive, arrays, and more tricky setups. For a couple decades I think there's been some good progress, but I don't think anybody has hit the homerun just yet with workflow or execution.
@ID_AA_Carmack I remember Nokia had its own version of this it was called OZO ahead of its time but suffered from the same fundamental issues. The hardware was impressive but the software pipeline (occlusion/stitching) problems were not solved at scale.
@ID_AA_Carmack ML is absolutely a wonderful tool to bridge the last mile between “all the data we can realistically capture and truthfully model” and “where we need to be” but the problem is that it’s being used as a means of “spend as little time and money to gather as little data as we can
@ID_AA_Carmack It amazes me that my employer could render full 3d scenes from lidar data clouds on standard computers, locally, in 1999. Im talking 800mhz machines with 128mb of ram, and IIRC, a Rage2 video card. That included point cloud registration for multiple scans, accurate enough to
@ID_AA_Carmack It's fascinating to see the innovative approaches towards spatial video capture—technology is truly pushing boundaries.
@ID_AA_Carmack i think this is okay to say publicly, but i interviewed a guy from meta around that time period who was working on the board designs for the prototype rigs for this sort stuff. 200+ spatially oriented cameras on a single board iirc to try to figure out which ones could be
@ID_AA_Carmack Vision hurdles.
@ID_AA_Carmack Did some experiments on this a while back (2015-2017). The issue for me was the data processing. Had some amazing rigs and ability to create true 3d captures of people/environements. Despite this, Gaussian splatting proves to be a much better solution simply because sharing a
@ID_AA_Carmack Spatial video evolution is fascinating.
@ID_AA_Carmack followed 📸
@ID_AA_Carmack Impressive depth, challenges persist. Excited for immersive media evolution.
@ID_AA_Carmack Ah the good ol’ days 😃
@ID_AA_Carmack Good times indeed
@ID_AA_Carmack @ID_AA_Carmack I think it's utterly hilarious that you babble and burble about "computer vision" and never even bother with things like IR optical materials or vacuum UV... ...unless! Unless, Carmack. You are not a smart person, Carmack, but you're required to play the part.
@ID_AA_Carmack Those rigs are wild! Reminds me of @RamseyJared94’s breakdown of how some indie filmmakers are hacking together cheap spatial video setups. Wonder if they’re using old-school CV tricks too.
@ID_AA_Carmack So what the price would look like and what countries can be supplies so ppl can access it? If not in Belarus or Russia then idk
@ID_AA_Carmack @grok is he moving from supporting raw computer vision to computer vison with the help of using AI ?
@ID_AA_Carmack What about light field cameras, aren't they supposed to solve this exact problem?
@ID_AA_Carmack Meanwhile, human eyes have a big ol' blind spot right in the center of each eye and the vast majority of humans don't even realize it.
@ID_AA_Carmack There were companies in the past that were pushing this envelope and still are, and they were not Meta and the like - @Mosaic360camera is the latest effort : https://www.mosaic51.com/, by Jeffrey Martin, he is in 360 imagery since early 2000s and did a lot in that field.
@ID_AA_Carmack Jaunt was a fun exercise for a while and we did a lot of work with them and engineering, the Meta/Red volumetric cam was interesting and we wrote a white paper on that for Meta. All the conclusions were exactly the same as yours @ID_AA_Carmack , close, needs work, too many holes
@ID_AA_Carmack i wish you were a billionaire instead of elongated muskrat
@ID_AA_Carmack So does the AI alone make a better job already than conventional CV algorithms? Or is the hybrid CV/ML required to obtain paramount result?





