Determining the resolution of the StarVR One [Nerd Talk]

I think your issue is just the low resolution spread over a very wide FOV. 1830x1464 per eye over 210 Degrees H is an incredibly low PPI. Almost exactly Oculus DK2.

It would be interesting to run some comparisons on resolution per inch vs. Index, Vive Pro, O+, 8KX etc… but this has to vastly limit the use cases for this device. My guess it’s vastly lower than anything else on the market even older headsets in that regard.

Yet, in terms of vertical resolution, that is almost the same as the Pimax 5k+ (1440v vs 1464v). So, intriguingly, the StarVR has a larger vertical FOV, or more pixels are discarded on that axis. Both at the same time would not surprise me.

From my spreadsheet work, I would guess the StarVR falls to the very bottom of a 40% total margin of error in estimating angular resolution based on vertical resolution and vertical FOV claims. And yes, that is very low resolution, close to a DK2.

Also, in hindsight, going back all the way to the DK2, it is amazing just how much of a difference there has been in optical clarity at each step from 1080v, 1200v, 1440v, 1600v, 2160v.

I think the importance of even small differences in linear angular resolution are greatly underappreciated by many.

2 Likes

Seems to be a calibration issue. At least it seems so after Sebastian got some file that I did not. Once I get those files I will know for sure

3 Likes

Not a lot calibration can do about such a low PPI. Might be able to correct various optical abnormalities though that occur from their Fresnel layering process.

Doubtful is a resolution matter. Before being resolution related, it would be an optics issue.

1 Like

They may have been forced to integrate a diffusion layer because spreading so few pixels over such a wide FOV would greatly exaggerate the gaps. I imagine it was very difficult to arrive at a happy medium.

3 Likes

This would be more or less what I have been thinking until Sebastian said that he got some files that neither me and John got that fixed the image calibration problem. I am still not sure if he is just blowing hot air or being truthful. I will wait until StarVR respond to my email about the matter about this calibration files

3 Likes

How do you calculate that? If I compare it to the Vive Pro:

  • Vive pro has 9.2M subpixels. However (roughly) 50% of that is ‘lost’ because it’s needed for binocular overlap. So that would leave 4.6M over 110 degrees FoV
  • StarVR has 16.1M subpixels. Only (roughly) 25% is ‘lost’ due binocular overlap so 12.1M is left for 210 degrees FoV

So it could offer a more detailed experience than the Vive Pro in my opinion.

This is based on a few assumptions, such that the Vive Pro binocular overlap is almost the whole FoV. It might be a bit less. Also they actually might have improved the binocular overlap for the starVR, so it might be a bit more than 25% that’s lost. Either way, I think it might be pretty close to a Vive Pro. Which would not be bad at all.

The biggest downside for the Vive Pro for me personally is not the resolution, but the SDE. However the StarVR has custom designed panels that are optimized in order to have a very high fill factor. So the SDE should be much better than the Vive Pro.

Also @mixedrealityTV said that the resolution seems to be close to the Vive Pro. So from the math above, I guess that makes sense

5 Likes

Wow those are some bad calculations and just wrong information.

First here is the famous Doc-OK’s 2019 paper on Vive Pro overlap:
http://doc-ok.org/?p=1955 (includes extremely detailed measurements using special equipment)
FYI it’s ~82%. If StarVR is anywhere near that their PPD is just 8.71. 8.71! (a pimax 5k+ is 21.33 for reference and the 8kX is roughly 37).
The human eye average overlap is just ~59%.

The calcs are on Wikipedia here: https://en.wikipedia.org/wiki/Comparison_of_virtual_reality_headsets
(they also calc the StarVR to be 8.71 PPD)

I remember reading somewhere the StarVR being 16 ppd.

I’d love to see how you arrive at those numbers.

2 Likes

These calculations for PPDs are wrong. I am pretty sure that Pimax 5k+ is ~ 14 PPD (Pixel per degree (PPD) of Pimax 5k+). OG Vive and Oculus have been measured by doc-ok differently too (http://doc-ok.org/?p=1694, Vive ~ 11.5 PPD, Rift ~ 13.5 PPD). I would not use the arbitrary numbers someone put there, without verification.

PimaxUSA Are you using PPD on a Pimax headset set to large FOV that Pimax recommends to owners they don’t use vs StarVR where the Large FOV is the FOV people will actually use?

It’s like arguing my car has 1000bhp but only two wheels v your car with half the BHP but 4 wheels you can actually drive.

Index doesn’t have the highest resolution or the widest FOV but combined gives you the best user experience at an affordable price.

Who will be first - Pimax and perfect lenses or StarVR with higher resolution screens?

1 Like

Can you explain what was badly calculated? As I said, I ASSUMED near 100% overlap for the Vive Pro (since the original Vive og supposedly was near 100%). I don’t know if it’s really 82% for the Vive pro, but if it is, I specifically said that it could be lower. That’s actually how I ended up to ‘similar as Vive Pro’, since if we’d take the 100% we’d come to almost 3x pixels over only 2x the FoV

Can you show us that calculation?

You can read through it on Wikipedia.

What exactly do you want me to read through? The wiki page you linked just shows some basis specs. You said I calculated things very badly, so I’d like to see your calculations. I mean if you say I did it incorrectly, at least show me what was incorrect and how it should be done.

4 Likes

I think some of the confusion here is between the apparent ‘roughness’ caused by gaps between pixels and such, and actual readability caused by not having enough pixels.

Apparent roughness can be satisfied with even the lowest capability headsets, like the Oculus CV1, as long as gaps between pixels are not too visible. Additionally, this may be ‘faked’ by adding diffusion, increasing overlap, etc. Which is why I keep telling people that for roomscale and such, the high resolution of the 8kX is not as amazing. It does make some things like vegetation in HL Alyx look better, but it’s not as much as a must have.

Actual resolution, cannot be faked with any magic. Only one thing counts - Pixels Per Degree (PPD). This is computed from exactly three numbers. Real pixels (not subpixels) along one axis, divided by field of view in degrees across that axis, multiplied by any concentration of pixels done by the lenses (so far this has been exactly 1 in all reported cases). Because these numbers have a ~40% margin of error due to such things as not all pixels being used, calculated PPD is a rough estimate. Thus, actual measurements and qualitative evaluations should be given considered as well.

Actual resolution, the 8kX is at the very top, and so it is the best for things like Virtual Desktop, where readability counts, not just smoothness.

@PimaxUSA The PPD numbers cited by that Wikipedia article, while resulting in similar relative rankings, in absolute terms, are consistently rather far from my own and some comparable estimates. IIRC, my own calculations place the 5k+ at 17 PPD and the Oculus CV1 at 11 PPD, not 21 or 12 and 9 respectively. Also, notice that two very different numbers are cited for the 5k+ depending on horizontal FOV setting. I think that article needs serious work on its PPD numbers.

@SweViver If your macro camera setup could be calibrated to exactly focus 10 degrees to a specified number of pixels on its sensor, then your Through-The-Lens photos could be used to put an exact number of VR headset pixels-per-degree count, ending the era of relative approximations.

EDIT: Additionally, at resolutions below a 5k+ at 1.75 Total SR, a lack of pixels per degree does create a severe ‘fog’ effect in any VR application. This is a sharp threshold, below which, an awful grey haze suddenly appears. Mathematically, it is due to the sudden loss of dynamic range as color information from multiple pixels is summed into a single pixel, resulting in colors not being as deep. Especially, the sharp transition was easily noticed in PavlovVR, which renders at very sharp contrast. PimaxUSA was definitely correct in calling this out.

1 Like

No. “real pixels” isn’t a good way to compare. If a headset only has 2 subpixels per pixel, then the effective resolution is lower than the specified resolution. In that case subpixels are shared between pixels, lowering the effective resolution. That’s the main reason (apart from the scaling of course) why the 8k barely looks better than the 5k+, the ‘effective’ resolution is much lower than the ‘specified’ 4k resolution due to this subpixel ‘sharing’

3 Likes

@Djonko

Usually, when I read ‘subpixels’, I think RGB. I see what you are referring to now.

The 8k+ is an unusual situation, using scaling. Since text is a bit more readable in the 8k+ than the 5k+, in fact all evidence shows the ‘shared’ pixels do not significcantly degrade visual quality, this data point does not change the validity of minimum visual quality estimations based on ‘real’ pixels. So until there is at least another case of this, which meaningfully changes the overall trend, I am sticking with real pixels in my discussions.

1 Like

But it does! How are you going to make a color if you don’t have RGB per pixel? The solution to this is that subpixels are being shared. So that the ‘grouped’ pixels create some average of what the original cluster would have been. This lowers the effective resolution. There’s a thread on this forum about it, I remember @jojon doing the most contributions there, I’m sure he can point you to it.

1 Like