VLC 3.0 - Blowing out bright areas of HDR10 content mastered at 4000 nits.

[CBme]

Giving the newly released 3.0 a spin and pleased to see how it leverages hardware acceleration to breeze though the mkvs form my UHD Blu-rays. While the colors now appear to be more accurately mapping to my rec 709, SDR PC monitor, I noticed on some titles the bright areas were massively blown out vs. what i see with MPC-BE w/madvr. The only pattern I can see after going through a few titles is that it only seems to have a significant difference on titles that are mastered at 4000 nits, not those mastered at 1000.
*EDIT. Going through more titles I do see a difference with 1000 nit titles too but not about them being blown out as bad (for those I've checked..) as the 4000 nit ones. Rather they are darker (than MPC) and the colors notably less vibriant. It is not the same 'washed-out'/muddy/flat situation it was before but it is not correct.

Is there a setting here (that I obviously can't find) that would address the tone mapping? Glad to submit bug report if that is appropriate.

https://imgur.com/a/Sgl0v

[Jean-Baptiste Kempf]

Are you using Direct3D11 or OpenGL for the video output?

[CBme]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

[agsmzt]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

New here but I solve this problem thanks to Jean-Baptiste Kempf idea.

Just setting Direct3D9 video output.

[Jean-Baptiste Kempf]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

New here but I solve this problem thanks to Jean-Baptiste Kempf idea.

Just setting Direct3D9 video output.

D3D9 does not show any HDR at all.

[Jean-Baptiste Kempf]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

The tone-mapping between dx11 and OpenGL is a bit different, indeed. And with a full HDR screen, it is even more different.

I am not surprised the HDR is not perfect yet in 3.0. MPC is good, but it's not always correct either. 2.2.x was catastrophic, 3.0.x is better.

So, could you:
- share logs?
- tell me what content did you try? The UHD 4K Spiderman and GoTG (the second one?), right?
- test with OpenGL with slow speed (should work-around the CPU consumption.

[agsmzt]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

New here but I solve this problem thanks to Jean-Baptiste Kempf idea.

Just setting Direct3D9 video output.

D3D9 does not show any HDR at all.

I know but my problem was that I cannot watch any video at all because so much bright.

Setting Direct3D9 "solves" my problem and I thank you.

[agsmzt]

Just take a look

Automatic Output
https://drive.google.com/open?id=19INkc ... TLurTH8waS

D39 Output
https://drive.google.com/open?id=1PIO6g ... uwmtL8ktF-

[CBme]

Just take a look

Automatic Output
https://drive.google.com/open?id=19INkc ... TLurTH8waS

D39 Output
https://drive.google.com/open?id=1PIO6g ... uwmtL8ktF-

That is a HD (ish) SDR youtube video from what I can tell, not a UHD (3840x2160, 10bit rec 2020, HDR10) video. Whatever issue is, it appears to be unrelated to to the issue of tone mapping UHD HDR content to SDR displays.

[CBme]

Are you using Direct3D11 or OpenGL for the video output?

It was set to automatic. Set it to dx11 and it performed the same. Set to opengl and it wouldn't use hardware decode so it couldn't be played.
Are you having different results?

The tone-mapping between dx11 and OpenGL is a bit different, indeed. And with a full HDR screen, it is even more different.

I am not surprised the HDR is not perfect yet in 3.0. MPC is good, but it's not always correct either. 2.2.x was catastrophic, 3.0.x is better.

So, could you:
- share logs?
- tell me what content did you try? The UHD 4K Spiderman and GoTG (the second one?), right?
- test with OpenGL with slow speed (should work-around the CPU consumption.

-Yes, glad to share whatever could be helpful. I don't know how to save a log in VLC at the moment but will figure it out. Which content do you want me to play and give you a log on? (and how should I get it to you as I don't see a way to attach things in this forum).
-Yes, those are the two examples but I tried about 5-10 others though I don't recall offhand which titles. I have about 25 UHD discs (all saved as movie+lossless .mkv, on my media server) and can try any combination of them based on attribute or title you'd like.
- I did so with the Spider-man Homecoming disc. Slowing it down did not work around the lack of hardware acceleration (with my Nvidea 1030) as there are apparently additional issues since it would only play a few frames and the video would never play a new frame though the audio would continue (however choppy). So I just found a random spot that works and matched up the video. Open GL for Windows doesn't have the blown out issue nor does it look like the video pre-3.0. More like what the 1000 nit mastered content does.

Matrix coefficients : BT.2020 non-constant
Original source medium : Blu-ray
Mastering display color primaries : R: x=0.680000 y=0.320000, G: x=0.265000 y=0.690000, B: x=0.150000 y=0.060000, White point: x=0.312700 y=0.329000
Mastering display luminance : min: 0.0050 cd/m2, max: 4000.0000 cd/m2
Maximum Content Light Level : 4332 cd/m2
Maximum Frame-Average Light Level : 992 cd/m2

[robUx4]

I made some changes to the way we process the tone mapping. The next nightly build (of VLC 4.0) should fix these issues.

[Jean-Baptiste Kempf]

not work around the lack of hardware acceleration (with my Nvidea 1030) as there are apparently additional issues since it would only play a few frames and the video would never play a new frame though the audio would continue (however choppy). So I just found a random spot that works and matched up the video. Open GL for Windows doesn't have the blown out issue nor does it look like the video pre-3.0.

I'd say the OpenGL display is pretty good, no?

[CBme]

I made some changes to the way we process the tone mapping. The next nightly build (of VLC 4.0) should fix these issues.

Great, I'll try to check it out sometime this week and report back. Please let me (us) know if there is a more useful or appropriate way to give feedback on the tone mapping to SDR topic than posting it here.

[CBme]

not work around the lack of hardware acceleration (with my Nvidea 1030) as there are apparently additional issues since it would only play a few frames and the video would never play a new frame though the audio would continue (however choppy). So I just found a random spot that works and matched up the video. Open GL for Windows doesn't have the blown out issue nor does it look like the video pre-3.0.

I'd say the OpenGL display is pretty good, no?

Certainly better than what is going on with automatic/DX11 in this example, without question. But as it won't play on my system, it is only really better if I wanted to look at a static frame.
That said, it does appear to be much darker than the default MPC+MadVR rendering which is itself darker than the standard rec 709 blu-ray displays (not sure which is more accurate to the source as I have UHD content but am still holding off on pulling a trigger on a UHD TV given the lack of HDR maturity and rapidly increase peak brightness capabilities). Also unsure as to whether OpenGL for Windows is suffering the same muted/darker color issue as DX11 is with the GotG since my system can't test it. If brand new powerful CPU's can power through OpenGL render without graphic card acceleration (of if the integrated Intel GPU's will do it) I'm hoping folks with the more modern processor can give it a go with their UHD rips to give better feedback than I can.

[Jean-Baptiste Kempf]

That said, it does appear to be much darker than the default MPC+MadVR rendering which is itself darker than the standard rec 709 blu-ray displays.

Unfortunately, this is quite hard to know what is the correct output...

So, for now, the goal is to get the same output as OpenGL with D3D11. And fix GotG.

Also you can try D3D9 + OpenGL.

[Jean-Baptiste Kempf]

Also, please test https://nightlies.videolan.org/build/wi ... 0213-0319/

[CBme]

Also, please test https://nightlies.videolan.org/build/wi ... 0213-0319/

I downloaded the latest, 4.0 - Feb 16th 64 bit, and the blown out images seen on the Spider-man example are gone. Yay!
However the muted colors issue remains with both the 4000 nit max and 1000 nit max examples I've looked at. Not the absurd level how it looked before it could do any tone mapping but notable enough that you wouldn't watch it this way if you had the choice.
Let me know if I can help any further. Otherwise I'll just check back from time to time. Thanks!

[BlueTemplar]

Hello !

I've done some tests that might be relevant to this discussion,
see them here :
https://arstechnica.com/gaming/2018/02/ ... t=34825757
P.S.: There are multiple posts.