Have we reached the limit for Quest2?

LordCrash

bobbytables if POVR ever figures out how to make an actual decent platform with 80% of the feature set from DeoVR

It plays VR porn pretty nicely.

It can be used with HeresphereVR and XBVR which is better than DeoVR anyway.

What else do you really need?

Yeah.

doublevr

Hurto11 yes, 8K is maximum for Quest 2 https://www.sexlikereal.com/blog/67-vr-videos-explained-4k5k6k-fov-fps-decoding

It's also true that higher resolution downscaled to lower will be better quality than lower resolution. That's why we are building 10k camera.

fenderwq

Rakly3 interesting. Will have to test playing two 8k files at the same time some time.

BTW, have you ever considered / thought about the following?:
So currently the max decoding dimensions for a h265 video file is 8192x8192 at 1:1. However, since vr is shot in 2:1 scale (meaning that max 8k would be : 8192x4096) you only utilize half of the pixels. So what if you had a custom lens profile that has double the pixels vertically (e.g. double the resolution vertically by squashing the image vertically on a larger texture)?
An alternative would be to have a different type of projection all together or at least "cut it up" in such a way that the full 1:1 or 8192x8192px is utilized (not sure how that last one would work though 🙂).
Seems to me that could potentially allow you guys to DOUBLE the resolution, which is pretty huge I think.
Anyways, I'm pretty curious if you've considered this and what your thoughts are

Rakly3

fenderwq However, since vr is shot in 2:1 scale (meaning that max 8k would be : 8192x4096) you only utilize half of the pixels. So what if you had a custom lens profile that has double the pixels vertically (e.g. double the resolution vertically by squashing the image vertically on a larger texture)?

Lenses don't have pixels. But I understand what you mean. You're describing for a large part fisheye lenses. The reason SLR Originals use the fisheye profile is because the closer to the center you look, the higher the pixel density is of the image.

What you are proposing would be like a barrel lens, but horizontally. Or maybe better, fisheye + barrel would be some sort of oval shaped lens. I see it as technically possible. But I don't think headset manufacturers would go for it as it would mean they have to completely redesign their software and lenses too. It's a good idea though!

fenderwq An alternative would be to have a different type of projection all together or at least "cut it up" in such a way that the full 1:1 or 8192x8192px is utilized (not sure how that last one would work though 🙂).

Yes, that is Viewport. Still would need new hardware though, or more than one decoder. Viewport 5K doesn't have the hardware limitation.

fenderwq (or maybe 4 if nvidia brings the argument that higher than 8k resolution has no practical purpose again).

Technically Nvidia is right. it's a lot easier to do parallel decoding of two 8K streams since it doesn't need any R&D. They already can do it.
Consumer grade Nvidia GPU's have 2 decoding streams (artificially limited)
Their professional GPU cards are 4 streams or 'unlimited'. The unlimited is a whole other topic though. They can make clusters of thousands of GPU's.

Blockchain also can make it possible with distributed computing (though pointless for decoding). But there are already projects in the works for this for encoding.

One of our devs even already ran a test project with one of my crypto mining rigs to upscale and encode to 16k by splitting the work over multiple GPU's. (This doesn't mean we will be having 16k video, it's for something else; SLR is more than just a studio 😉 )

Sandi_SLR

fenderwq Hey if this is true we can cut it like this:

size of the fisheye if 8192x4096
size of the fisheye if 8192x8192
how to pack it (this would make the diameter 73% of the resolution, so roughly 5980px square per eye)

fenderwq

Rakly3 your remark of playing two files at the same time got me thinking. One would probably be able to decode at least 2 times 5790x5790 or 11580x5790 total as it's the same amount of MP as 8192x8192.
That's seems very interesting to me also, especially considering you have a 10k camera coming up.
I'm sure you thought about these kind of things as well but I'm just really curious about what will be possible the next few years. It would just be amazing if we could surpass that pesky 8k limit the next two years (or maybe 4 if nvidia brings the argument that higher than 8k resolution has no practical purpose again).

Rakly3

fenderwq One would probably be able to decode at least 2 times 5790x5790 or 11580x5790 total as it's the same amount of MP as 8192x8192.

There are dimensional limitations too, and not just the MP. I don't know them off the top of my head, as this also depends on the codec. But let's assume we are talking about mpeg-4 group H.264 & H.265
(don't confuse with mp4 container file format.)

Slicing up the image as proposed by @Sandi_SLR shouldn't cause issues with displaying the image correctly for a computer; There is no transformation of the image, just location of the pixels. It's no more difficult as playing video in mirror or upside-down, equirect, fisheye, cubemap.. - Textures on 3D models, and cubemaps in a game are way more complex than that and done in real time.
VR on YouTube already does this. Image search ' youtube cube wrap vr '

The dimensional limitation comes from how many slices can be used and their maximum size. Known as Macroblocks.
In H264/AVC they are straight forward (doesn't mean it's easy material), but become a lot more complex in H265/HEVC.
https://en.wikipedia.org/wiki/Macroblock
Ever seen blocks in an image, or wonder why artifacts are always square-shaped? (Rhetorical)
https://en.wikipedia.org/wiki/Compression_artifact

These macroblocks are also where the max resolution comes from.

bobbytables

Rakly3 Lenses don't have pixels. But I understand what you mean. You're describing for a large part fisheye lenses. The reason SLR Originals use the fisheye profile is because the closer to the center you look, the higher the pixel density is of the image.

true, although if someone wanted to be cheeky they could bring up that lenses do have a maximum resolving power and that's usually measured in pixels, but that doesn't really have anything to do with what you're talking about

fenderwq

Rakly3 Cool, thanks for the reply, can't help but love this kind of tech talk 🙂 Some really cool ideas there, didn't know about the parallel decoding either. Gives me hope that we don't have to wait for 8k+ for years to come!

Guess I skipped a few steps in explaining what I meant with a "lens profile" though. To clarify, what I meant is that you convert the footage of two cameras (one for each eye) that have higher than 4k resolution to a single elongated (fisheye) side by side output movie and then in the software tell it to interpret it as if it was shot with a lens that has twice the pixels vertically. Similarly to the way you have to interpret fisheye content differently as you would equirectangular content in DeoVR (this is what I call a lens profile btw but maybe that's incorrect 🙂). This way you would still be inside the resolution limits of the decoder but also have twice the vertical data for rendering the texture (theoretically at least and with 2 8k camera's).

The second one was indeed what Sandi_SLR showed (thanks for the cool visualization btw). Correct me if I'm wrong but once you decode this 8192x8192 image it's up to different parts of the GPU to map it to a texture and display it in the headset right? So in theory you should then be limited by the "overall power" of your gpu and no longer by the hardware decoder?

Anyways, however much I like to go on about this kind of tech talk, do you think it realistic that we could see something like 10k/12k or whatever during this generation of video cards? What are your thoughts?

Ventriloquist_Tacos

I certainly hope we don't have to wait for the RTX 5000 series or worse in 2024/2025 to at least start seeing some experimental 10k/12k files, file size be damned. 50 GB, load me up. As most in this thread probably already know, the current mid to high end headsets have already maxed these files out. However, this is going to get worse soon.

2023
Quest 3 - Rumored by SadlyItsBradley to have 30% higher resolution, which is consistent with past increases
Pimax Crystal - Insanely high 2880x2880 per eye resolution with QLED, quantum DOT layer and local dimming
Apple - late 2023, maybe? 4k per eye rumored

2024
Quest Pro 2 - per Meta's usual cadence (year unconfirmed but something either 2024 or 2025 is certain)
Other high end headsets - particulars are unconfirmed but as a law of averages we have to see something from at least some of the following: Pimax (12k), Valve, HP, Samsung, etc.

So for the love of all that is enthusiast VR porn, SLR or whoever please get us some 10k content. Even 10k at approximately 10,000 x 5,000 (50MP) would blow 8k out of the water, so this would hold us over a few years.

Rakly3

Ventriloquist_Tacos in VR video, the display pixel dimensions are not a measure of quality. Instead of PPI (pixels per inch), it's the PPD (pixels per degree) that count.

Pimax 8k, 12k, Apple 4k/eye etc do not tell you anything about image display quality.
pixels over FOV does.

8K over 180° FOV is the same PPD as 4K over 90° FOV, and 2K over 45° FOV
Other factors also play a role in quality, I just wanted to illustrate that you don't have to fixate too much on the 'K'. It's just a marketing ploy. You don't see them advertise the PPD very often.

Ventriloquist_Tacos

Rakly3 Hi Rakly, I know the resolution/PPI/PPD/FOV stuff have common misconceptions but I already understand the points you made. You may be able to help me understand something else, which I'll get to below.

My point was that as we continue to get higher quality displays (whether QLED or OLED) with better colors, and particular with higher resolutions, one way or the other, the limiting factor is going to be the video file rather than the display. I also understand that increasing resolution alone on video files is not enough if the bitrate is low, lighting is terrible, etc. etc. Speaking to PPD, we're going to see average VR headset PPD run from 20ish to 35 or 40, and later beyond. Here are my questions as I've been experimenting lately:

(1) Does zooming out in a video represent a true PPD increase in the video? For example, In HereSphere you can zoom out from 180 degrees all the way to 90. Now, 90 is not ideal because it makes the actors too small and distorts the image. But I will say that it clearly makes the image much sharper. Is this literally a doubling of the effect PPD when you do this? If so I'll probably opt for an in between setting, like 120 degrees to get a happy medium with actor size and higher PPD.

(2) PPD calculations are all over the place online, to the point I can't figure out how they are actually calculated. When people talk about the Quest 2 for example, they frequently refer to the PPD around 20-21. But when you divide the total horizontal pixels (3664) by 106 degree horizontal FOV you get 34.56 PPD. This however is Varjo Aero level, which doesn't make any sense. Then, take the Pimax Crystal's claimed 35 PPD lenses. Those lenses at 5,760 total are quoted to have a 140 degree horizontal FOV. But if you divide 140 into 5760 you don't get 35, but 41. If you do this for the 42 PPD lenses you get 53 (110 horizontal FOV). My point is people continually say PPD is calculated by horizontal FOV divided by degrees in the said FOV but I literally never come up with the same number and sometimes it's off by a huge amount, such as in the case of the Quest2.

https://www.reddit.com/r/virtualreality/comments/qgefk5/ppd_comparison_chart/

« Previous Page