Could it be the display id causing incorrect info? Being presented to steam causing resolution difference?
Or as someone asked about usuable res due to screen utilization?
I am asking btw. 
1 Like
That would be on par with what one of the Pimax guys said about the 5K+ having more usable pixels than the 8k (can’t find the quote right now, what was the percentage again ???). But then, the 8k will never display an image as crisp as the 5K+… we will always have a nice upscaled 4K picture for each eye, with very little SDE, but with a smaller source (pre-scaling) image, and therefore lost details.
1 Like
Supposed to be 9% improvement over reg 5k. Due to smaller panel.
The usuable pixels though that would be only in the input res?
1 Like
You are forgetting the other factor which may impact the equation - PiTool “Rendering Quality” option.
What can be done to try to figure out at least the relative difference between 5K+ and 8K resolutions:
- Set PiTool to 1.0 “rendering qualtiy”
- Set SteamVR supersampling to 51% (it is an arbitrary but fixed value, I just have good reason to try this one)
- Record what is SteamVR reported rendering resolution.
Repeat for the other headset.
Factoring the reported resolutions should give the relative difference of targeted resolutions of individual headsets, assuming that “rendering quality” set to 1.0 will expose their nominal values.
@SweViver, could you try that?
2 Likes
Ah ok, that’s compared to the regular 5k, not the 8k.
We definitely need some clarifications here, please, @deletedpimaxrep1
1 Like
I already asked for such a test, it would be awesome if one of the testers could do it !
1 Like
Now it would have a higher panel utilization though due to smaller panel. Ie 5k & 8k might have a 5" panel. Where as the 5k+ might have a 4.5" panel.
[quote=“Heliosurge, post:67, topic:8249, full:true”]
Now it would have a higher panel utilization though due to smaller panel. Ie 5k & 8k might have a 5" panel. Where as the 5k+ might have a 4.5" panel.
[/quote]Thing is, the precise answer to these questions would make the 5K+ definitely the best headset unless you don’t mind a lower resolution blurry picture… or not if it’s just a bug somewhere.
1 Like
Nice to see someone else trying to approach this using information theory.
While you are right, the 8K image contains more basic information than the 5K image (that the upscaler adds), it actually contains less GPU-produced image information (because 4k is not an integer multiple of 1440p) it cannot be perfectly mapped to the display, so some GPU information unavoidably gets lost, as the upscaler can only ever approximate it not reproduce it perfectly accurately.
It’s like taking an email, replacing every 3rd word with any 2 random words that would still fit grammatically. Yes there are now more words in total, (so at the pure Infomation Theory meaning it now contains more information), however the key part here is that introducing those extra guessed words adds nothing to your understanding/perception of the original message (i.the original image as rendered by the GPU), and may even obscure it.
I’m perfectly aware of all that you’re saying. I’d be a very bad graphic software engineer if I didn’t.
And your email analogy isn’t exactly correct. A picture isn’t words, and it’s based on only 4 data per pixel, red, green, blue and alpha (transparency).
Image information doesn’t get lost in an upscaler, only transformed (unless the upscaler really sucks, but not even the simplest algorithm will do that). Of course, the result will be more blurry than the original, because there’s only so much you can “guess” from the neighboring pixels, but for such a small factor as in the case of the 8K, it shouldn’t be such a big issue, specially not for text, which should be just as readable as on the 5K+. That’s why I suspect there’s something else in play here.
Image information gets lost when the resolution is reduced. Then pixels disappear that will not be able to be “guessed” back. That’s what happens in the last picture.
3 Likes
But has anyone complained about the lack of clarity in the 4K?
No a picture isn’t words, but on an Information Theory level as we were discussing, pixels and words are all just information. That is fundamentally true regardless of the system of encoding used to represent them.
You’re quite wrong about upscaling never loosing image information. Discrete (i.e. digital not analog-based) upscaling absolutely can’t avoid loosing information whenever the scale factor isn’t an integer multiple. i.e when either dimension of the output resolution is not an integer multiple of the input signal’s resolution (as is the case with the 8K), as you will inevitably get some “averaging” across pixel boundaries.
I do agree that the “guess” factor of the 8K is relatively small (and that there appears to be something else going on too) but it’s still not zero like it is with the 5K+.
[quote=“JustNiz, post:72, topic:8249”]
Discrete (i.e. digital not analog-based) upscaling absolutely can’t avoid loosing information whenever the scale factor isn’t an integer multiple.
[/quote]Transformed, not lost. It’s just represented in a different way. The biggest issue with basic upscaling algorithms is the loss of contrast on small details, because something like e.g. a hair, 1 pixel wide in the original, will be dispatched in several neighboring pixels in the end result. You can clearly see that in the third picture I posted. And it happens with any upscaling factor, the bigger, the worse of course.
Loss = some data is not represented in the result. That’s when you downscale a picture.
Anyway, that’s not really the point to be made here. Considering the modest upscaling factor of the 8K compared to the original picture, it’s strange that there are so many differences between the result on the 5K+ and the 8K.
To make it short, in two sentences:
- The problem of upscaling is that your are limited to the information of the smaller source - no loss, but spreading the lack of information.
- The problem of downscaling is that you are reducing the information of the larger source - loss of information.
So when the original image is first downscaled to fit the input picture of the device, and then upscaled back to the real resolution, the result isn’t very nice looking… because first you get rid of information, and then you spread out that loss.
1 Like
In your email analogy you have to add a letter every two, not replace it.
upscaler = upqscdalmers
absolutely less readable, but no original information is really lost
2 Likes
Transformed, not lost.
Information is definitely lost even in your example. A 1 pixel wide entity will loose its edge definition when scaled up using a non-integer ratio. for example say it scaled up to 2.2 pixels, even assuming it happened to fall perfectly aligned on the “left” pixel boundary so it had one “hard” edge, the right side would only have (0.2 * 100) i.e. 20% of an overall effect on the actual colour of the 3rd pixel, so would appear blurry. In Information Theory terms, that bluriness is due to induced entropy, and no longer shows the hard edge actually present in the original image, so absolutely is information loss.
1 Like
I don’t think you understand the maths behind this, and his has nothing to do with integer or non integer ratio. All data in transferred to the new, bigger bitmap. All the data of what you call a “hard edge” is actually still present spread into several pixels instead of just one. There’s no “entropy” introduced except of course if you use stuff like fractal algorithms. if 0.2 if a pixel goes into one, the 0.8 will go in others. Just spread out. Unless your upscaling algorithm really sucks badly of course.
But whatever… the original point stands. No way the 8K image should be so much worse than the 5K+ unless there’s something fishy we don’t know about.
I don’t think you understand the maths behind this,
Lol I have an honours degree in computer science with a speciality in information theory, so I kinda think I do, especially on trivial stuff like this.
his has nothing to do with integer or non integer ratio.
Yes it really does. The only way to scale a digital image losslessly is to do it with an integer scaling ratio, otherwise you will always get some level of discrete quantization. It’s exactly the same reason that digital computers can’t represent a floating point number with infinite precision, and that digital music formats can’t ever perfectly store analog sounds.
All the data of what you call a “hard edge” is actually still present spread into several pixels instead of just one.
…and in spreading it, it gets mixed with other data so is effectively lost, as you can’t reconstruct which part came from where, especially with a display that uses a finite number of pixels.
There’s no “entropy” introduced except of course if you use stuff like fractal algorithms.
Sorry but very wrong. And fractals have exactly nothing to do with it. Its entropy through quantization. Here go read up on it: Quantization (signal processing) - Wikipedia
5 Likes
Vive at 110 is measured diagonally. StarVR is measured horizontally. Pimax is measured diagonally like Vive.
I definately like the 8K more. I truly hate SDE.
2 Likes
DO you have a credible link that says the vive is definitely measured diagonally? I went looking for exactly this a while back and couldn’t find anything definitive. Actually.I found a lot of stuff that seemed to say it was vertical (!).