Recommended settings for 10 bit SDR (Nvidia Shield)? #746
-
|
I'm using an Nvidia Shield 2017 (HDR10, no DV) with an EZ-SP12H2 splitter, a MS2130 (old YUV firmware), and HyperHDR 20.0.0.0beta1. The problem I experience is that SDR content is captured with a low dynamic range (e.g., the lowest black level in a live video screenshot is RGB 30,30,30), and thus the backlights are generally too bright for dark areas and too dim for bright areas. I assume this is related to the video output mode of the Shield, which is a Are there any recommended settings to get the SDR image processed correctly? Forcing the Shield to output an 8 bit signal works, however, this disables HDR playback and is thus not a viable solution. |
Beta Was this translation helpful? Give feedback.
Answered by
awawa-dev
Jan 25, 2024
Replies: 5 comments 33 replies
-
|
Hi LUTs available for download assuming the best full color range, if not you will need to calibrate the LUT yourself. Calibration on your own is always the best option, because it takes into account the setup you use every day. You should perform calibration for the (limited) video mode you are using on shield (best option or PC): that should do the trick both for SDR and HDR. If you want to have better HDR colors then you can check new feature to calibrate it better: #499 (comment) Installers: https://github.com/awawa-dev/HyperHDR/actions/runs/7409204606 (branch: 422_limited_calibrator) Full calibration logs appreciated. |
Beta Was this translation helpful? Give feedback.
5 replies
-
|
Thanks. I performed the calibration with a browser on the Shield, and the lowest SDR black level is now 15,15,15. Is that expected, or should it be 0,0,0? Btw. most Android TV browser apps don't offer a full-screen mode and thus don't allow running the calibration (error: |
Beta Was this translation helpful? Give feedback.
-
|
All information is available in the calibration logs - without them it is just empty speculation on how it currently works. You can perform two calibrations (first set the Shield to SDR mode and then to HDR mode: the calibrator supports both modes) to get more information about the range of values of each signal. |
Beta Was this translation helpful? Give feedback.
-
|
The logs say Full logsFor comparison, putting the Shield into Full calibration logs (8 bit) |
Beta Was this translation helpful? Give feedback.
-
|
Thank you for the logs, they were useful. |
Beta Was this translation helpful? Give feedback.
-
|
Thank you, that has fixed the black level. The white level is still at |
Beta Was this translation helpful? Give feedback.
-
|
The problem with your signal (and what I manage to reproduce on my A ms2130 grabber with old firmware and Ezcoo splitter) is this value in SDR (YUV 10bit limited):
This does not correspond to any standard, e.g. 16-235. I make some adjustments to try to take this difference into account, but the results may vary. In addition, I don't know if I want to and maybe remove this LUT table because... I noticed on my second B ms2130 grabber (where I uploaded the old firmware + experimental EDID) that for the same GFX settings, instead of the range 29-219, we get the nice standard range 16-235. I don't have NVidia Shield and I don't know if the firmware is the solution for this case also. So it turns out that this is the second serious problem with the original firmware: first a sporadic, very dark screen in cooperation with Ezcoo and now an strange narrower range when GFX is using YUV in the limited range. |
Beta Was this translation helpful? Give feedback.
9 replies
-
|
I think I have trouble understanding how to correctly calibrate. Did I understand correctly, that the calibration needs to be run twice: once for SDR, and once for HDR? If so, how do I tell HyperHDR which mode I'm calibrating? |
Beta Was this translation helpful? Give feedback.
-
|
We usually do not need to perform SDR calibration: here we did it only for diagnostic purposes to determine the operating ranges of the grabber, which allowed us to detect the problem with a very narrow range of 29-219. The most critical part for SDR is detecting whether the YUV is full or limited. You have this information in the final report: Theoretically, it is also possible to combine an SDR calibrated LUT with an HDR calibrated LUT into one. Each LUT file consists of three blocks of 256 * 256 * 256 * 3 (50331648) bytes each: you need to take the third block from the SDR LUT file and paste it at the same offset into the HDR LUT file. |
Beta Was this translation helpful? Give feedback.
-
Thanks, I got totally confused about Shield SDR vs HDR output range (8 vs 10 bit), and HyperHDR SDR and HDR processing. So what I posted above was an HDR calibration after all. |
Beta Was this translation helpful? Give feedback.
-
|
Do you have logs? 420 10 or 12 bits HDR was used? |
Beta Was this translation helpful? Give feedback.
-
|
It's the ones posted above in #746 (reply in thread) |
Beta Was this translation helpful? Give feedback.
-
|
@protyposis That's excellent news! Thanks for testing. I have the last request for you because I updated ms2130 LUT for limited yuv modes on HyperHDR website (what you tested previously, installation from HyperHDR). Could you test it?
I am attaching the calibration result of this LUT for (GFX: YUV 12bit limited, HDR, ms2130 is using our f3 firmware) . |
Beta Was this translation helpful? Give feedback.
13 replies
-
|
BTW if you look into HyperHDR code then again in your logs: |
Beta Was this translation helpful? Give feedback.
-
|
lol I'm so sorry for this confusion, but I think it is slowly clearing up. To reiterate, the Shield always outputs YUV 10 bit. (I can force "true SDR" 8 bit RGB or YUV output, but then it cannot display HDR content at all beause it cannot automatically switch between 8 bit and 10 bit output.) With 10 bit always on, the only difference between SDR and HDR playback is the color space it uses: Rec. 709 vs. Rec. 2020. All my previous calibrations were made with Rec. 709 output. I think that is what you referred to as SDR, while I (wrongly) called it HDR. It is possible to force the Shield to output So the mean error became very high, while it still says Full logs (YUV 10 bit Rec. 2020) |
Beta Was this translation helpful? Give feedback.
-
|
Ok, I think I understand what's going on too. SDR was simply accepted as the least evil but the calibration failed. I haven't seen the result.
Unfortunately, this is an almost unheard of possibility among players (I'm not sure if I've ever encountered it). And that's why I didn't take it into account in the calibration. It's a pity, because in this case the results of tone mapping would be even better, because the result is not packed into a narrower range. So to sum up: at least at the moment, this mode is not supported: neither by a pre-built LUT nor by calibration. |
Beta Was this translation helpful? Give feedback.
-
|
Cool, I'm glad we could clear this up :) |
Beta Was this translation helpful? Give feedback.
-
|
We concluded only that this mode is not-supported based on your report. |
Beta Was this translation helpful? Give feedback.
-
|
After extensive discussion, the current answer for Shield 2017's default output mode in a HDR setup (YUV 10 bit / HDR ready with "Match content colour space" enabled): it isn't fully supported by HyperHDR due to the rare output mode (#746 (reply in thread)). SDR works great, HDR not yet. The best configuration at this time is installing the |
Beta Was this translation helpful? Give feedback.
0 replies
-
And how do jump you to such conclusion? It is pure manipulation. because quote says:
|
Beta Was this translation helpful? Give feedback.
6 replies
-
|
I just reproduced what you concluded yourself:
Forcing the Shield to output Rec. 2020 is not an option because SDR content isn't rendered correctly in this mode. Similar to how 8-bit RGB/YUV output is not an option because it disables HDR playback. Also, there is no API on the Shield to programmatically switch the output mode, which could be combined with HyperHDR's API call to toggle HDR mode. So in summary, I don't see this as a viable option, but of course YMMV. |
Beta Was this translation helpful? Give feedback.
-
|
This time I will quote you twice:
vs
[logs show unsupported mode and based on this you reach manipulated conclusions] But in above operation you have performed such LUT calibration using Shield option that you do not want to use. However, if you want to add direct support for Rec.2020 coefs in HyperHDR and you also have a test environment, you can start with coefs: KR = 0.2627, KB = 0.0593. We appreciate active participation in the development of the open source project. Nevertheless, thank you for confirming on another platform that the ms2130 with the old firmware has a problem with the YUV limited signal (RGB range 29-219 instead of 16-235) and upgrading ms2130 to our experimental firmware (f3) solves it also for NVidia Shield. After a firmware update, 10-bit SDR is fully operational on the NVidia Shield using the Rec.709 standard. Rec2020 is not designed for SDR content, so at least for now it has not been considered nor do I have the setup to implement & test such a specific combination where it is forced. |
Beta Was this translation helpful? Give feedback.
Answer selected by
awawa-dev
-
|
The Shield uses Rec. 2020 for HDR content, and Rec. 709 for SDR content. So when I play SDR content with Rec. 2020, I get a dim picture because the color mapping is off. Hence, forcing Rec. 2020 is not a viable solution for SDR playback. |
Beta Was this translation helpful? Give feedback.
-
vs
So don't force Rec.2020 for SDR because neither Shield uses it for SDR nor Rec.2020 is implemented in HyperHDR (because it wasn't necessary before for such specific configuration) and they are waiting for you to contribute PR because I guess you need this functionality since you started this discussions, you have programming skills and working test configuration. You got the Rec.2020 coef parameters on a plate in my previous post. |
Beta Was this translation helpful? Give feedback.
-
|
Maybe you want to re-read the configuration suggestions I posted as the current solution to this discussion, because they already covered the color spaces: #746 (comment) |
Beta Was this translation helpful? Give feedback.
-
|
Unfortunately, as I described it in the above post, you are contradicting yourself. You had a chance now to address it and clarify it, but you didn't. Maybe it's a language barrier, but in its current state we won't go any further, especially if you've already resorted to manipulation, which is against the Code of Conduct. |
Beta Was this translation helpful? Give feedback.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
This time I will quote you twice:
vs
[logs show unsupported mode and based on this you reach manipulated conclusions]
But in above operation you have performed such LUT calibration using Shield option that you do not want to use.
And manipulated my quotes. I think that's enough. I least I have enough. And I dont have NVidia Shield to test it …