AVS Forum banner

1221 - 1240 of 1254 Posts

·
Registered
Joined
·
4,361 Posts
Discussion Starter #1,221
Thank you for your responses, and for the disc. I've been able to reproduce the situation using CalMAN and the VFPro, but that leads to additional questions on my part. Attached are two gamut measurement screenshots, both for P3D65|BT.2020 using the same metadata as on the disc, one with 25% sweeps at 50% stimulus, and one with 25% sweeps at 100% stimulus. (The version of CalMAN I'm using doesn't give the option for just 100% saturation for P3 for this page.) The 50% stimulus results are in line with my expectations, having taken similar measurements multiple times in the past. The 100% stimulus results were a surprise, however. The color patches that are displayed look the same as what I see with the disc, but CalMAN is telling me that what I see is largely what I should have in fact expected. The 100% measured values are certainly off from their reference points, with some consequently high dE's, but not half-a-gamut off. At 100% stimulus, the reference points (the small squares in the CIE chart) for 100% saturation P3 primaries have been pulled way in, such that red is pretty orange, green is basically yellow, and blue is pretty magenta. The screenshot has 100% green selected, and you can see by the zoom-in in the upper right of the CIE chart that the reference points for green and yellow are basically on top of each other. I assume this squishing is due to the volumetric shape of BT.2020? And also, when you say the P3 gamut patterns are at 100% stimulus, you mean 100% in BT.2020 terms, yes? So, does this clear anything up, or just add more confusion?
View attachment 3040487 View attachment 3040488
100% stimulus is 10000 nits in HDR. That is why we setup the saturation sweeps at 50%.


Sent from my iPhone using Tapatalk
 

·
Registered
Joined
·
3,180 Posts
Thank you for your responses, and for the disc. I've been able to reproduce the situation using CalMAN and the VFPro, but that leads to additional questions on my part. Attached are two gamut measurement screenshots, both for P3D65|BT.2020 using the same metadata as on the disc, one with 25% sweeps at 50% stimulus, and one with 25% sweeps at 100% stimulus. (The version of CalMAN I'm using doesn't give the option for just 100% saturation for P3 for this page.) The 50% stimulus results are in line with my expectations, having taken similar measurements multiple times in the past. The 100% stimulus results were a surprise, however. The color patches that are displayed look the same as what I see with the disc, but CalMAN is telling me that what I see is largely what I should have in fact expected. The 100% measured values are certainly off from their reference points, with some consequently high dE's, but not half-a-gamut off. At 100% stimulus, the reference points (the small squares in the CIE chart) for 100% saturation P3 primaries have been pulled way in, such that red is pretty orange, green is basically yellow, and blue is pretty magenta. The screenshot has 100% green selected, and you can see by the zoom-in in the upper right of the CIE chart that the reference points for green and yellow are basically on top of each other. I assume this squishing is due to the volumetric shape of BT.2020? And also, when you say the P3 gamut patterns are at 100% stimulus, you mean 100% in BT.2020 terms, yes? So, does this clear anything up, or just add more confusion?
View attachment 3040487 View attachment 3040488
What white point did you calibrate to?

Sent from my LGMP450 using Tapatalk
 

·
Registered
Joined
·
9,497 Posts
Yes to your questions. Calman should be able to generate the same patterns with the VF Pro. On the gamut, stimulus and saturation are at 100%. One is DCI-P3 mapped into 2020 and the other is pure 2020. For the saturation sweeps, stimulus is at 58% (Calman currently uses 50% and will change to 58% in a future release)
Hi Stacey,

Tyler can open the workflow, change to 58%, and save it, then include that workflow to the next CalMAN.

If you save the workflow, and next time you will open, it will have 58% as default as the settings per each layout page are stored to the workflow file.

CalMAN requires an update to support ColorChecker @ 58% in PQ.

For saturation measurements, within the gamut, then you use a lower stimulus. Where we use 58% and Calman uses 50%. The 50% was just a guess back in 2014 to figure out a good starting point for HDR and wide color gamut. The ITU and others later settled on 58%. I keep bugging Tyler to change it in Calman and one day he will get around to it. :) It is on his list. You can manually change in the UI, so not the end of the world.
When ITU's BT.2408 specifications released, defuse white became 203 nits, so 58% of the full PQ signal (input) level.

It was 100 nits in the past; its the reason CalMAN had it at 50%.
 

·
Registered
Joined
·
9,497 Posts
Change from CalMAN Targets that you have 10000 nits peak (not use measured White), this trick will fix CIE bug.
 

·
Registered
Joined
·
1,248 Posts
Change from CalMAN Targets that you have 10000 nits peak (not use measured White), this trick will fix CIE bug.
OK, thanks, attached is a screenshot with just that change, no new measurements. So the reference points using measured White are not to be trusted at all, or these are relative vs absolute views of the display's capabilities?
P3 100% 10K White.png
 

·
Registered
Joined
·
1,248 Posts
P3D65 Red (Title 72, Chapter 2) w/ Panasonic UI showing P3 metadata (Alters the color slightly while UI is up)
View attachment 3040433
Arghh, I was only half awake earlier today, and wasn't paying enough attention. If the gamut patterns are at 100% stimulus, MaxCLL should never claim to be only 1000 nits, right? If I play the patterns back with 10000 nit metadata, I get reasonable colors. Ditto if I generate patterns with the VFPro using 10000 nit metadata. My display is keying off that metadata (and perhaps doing a suboptimal job when the MaxCLL is incorrect).
 

·
Registered
Joined
·
5,580 Posts
Arghh, I was only half awake earlier today, and wasn't paying enough attention. If the gamut patterns are at 100% stimulus, MaxCLL should never claim to be only 1000 nits, right?
You can't trust MaxCLL. Some studios set it to 0 on purpose. Some will remove an outlier that might cause a movie that is under 300 nits to have a MaxCLL of 9000 and some will not. Some discs have MaxCLL values higher than Mastering Display Luminance Max value.

Some displays will use the MaxCLL value to change the tone mapping curves and some do not. We offer all the patterns at various levels to cover several scenarios, but there is no agreed upon standard on what level to use for HDR10. Dolby Vision solves this problem, so its not an issue if you set the proper bits.

The important thing for calibration is to use static metadata. We defaulted the disc to 1000 to match Calman. Calman set the default to 1000 because of the X300. You can make it anything you want in Calman, but a disc is limited. Each level is a different encoded video file.

I am glad we know where you saw the difference.
 

·
Registered
Joined
·
3 Posts
Recently, I have been out of the loop regarding Spears & Munsil HDR Benchmark disk new edition.
I went to the website, but I couldn't find any information on the subject.
I know it was a thread about it here but I can not find it.
Is there any news about the release time for the new disk?

Thanks in advance!
 

·
Registered
Joined
·
3,180 Posts
@sspears Well, I come to say thank you and your partners for the hard work and commitment to PQ excellence. The calibration done with the windows on your disc has change the game. It makes a strong case, for the continued support of physical media.

At this point, if the tech in my display would allow it. I don't think I need to upgrade my display, but I am a gamer so upgrading will have to happen.

Money well spent.

Sent from my LGMP450 using Tapatalk
 

·
Registered
Joined
·
5,580 Posts
See, for example, were stills from S&M UHD:

The out of gamut traces mainly come from the fact that the clip has values below 64. There frames with above 940 (10000 nits) also, until 20000 nits.

The in-picture frame overlay is the HDR Histogram, which shows the distribution of video content levels across the entire frame.
This is correct, we have values below 64 and above 940. These naturally occur when you convert from RGB to YCbCr. I have verified it occurs in our tools, Dolby's, ColorFront's and Resolve's. Some will chose to clamp and some will not.

In 2007, Ned Price would personally QC 5-minutes of every encode for HD DVD. Ned was VP of technology at WB. GDMX, which was owned by WB, did all of the encoding back then. GDMX, against my wishes, applied a clamp to Batman Begins and Constantine, which were the first two titles they submitted to Ned. Ned did not even look at the BVM, he notice on the WFM they had clamped and immediately rejected the encodes. He thought it was VC-1s fault at the time. GDMX re-encoded w/o the clamp, submitted and they passed.

I mention because the Dolby Vision version on the current disc was 100% controlled by Dolby while the rest were controlled by me. This means that Pixelogic, the premiere encoding house, encoded the Dolby Version. Neither version clamps the signal.

After seeing Ted's post, I did encode a version that was clamped and have since removed the clamp and started re-encoding and I don't believe clamping is the proper thing to do.

In the example photo from Ted, you can see one of the issues in the old grade that drives me crazy. The crane on the right has purple LEDs. They are actually blue. It is now proper blue in the new grade. It was a bug that has since been fixed.

These limits are taken automatically from the Mastering Display Luminance minimum and maximum values from the clip's HDR10 metadata.

So the levels shown with red color are brightness levels below and above Mastering Display Luminance values.

The specific HDR10 clip had 0.005 nit black (min) and 10000 nits (max) white.

The levels within the Mastering Display Luminance values have a green color.
This is a great topic, thank you for bringing up Ted.

Mastering Display min and max are not meant to mean nothing will be below or above. They are simply the capabilities of the mastering displays. The senior and master colorists I have worked with all do the same thing. They use a scope and set black at true black. So they will always go below 0.005 nits. Credits and letterbox bars are simple examples of what will go below 0.005 nits. I don't know if every colorist does this.

Dolby also asked me to not change the mastering display min from 0.005 for the montage, so we have left it that way on the new version too. The skin tone content has min set at 0.0001 since I used an X300 to QC. Technically the skin tone content was not graded. I white balanced it and then used Transkoder to convert the Log3G10 into 2084. Phil did an amazing job lighting that no grading was needed.
 

·
Registered
Joined
·
5,580 Posts
Just want to provide an update on our status.

We started encoding all of the HDR patterns on 8/29. We had to start over twice, the last time was two weeks ago. The first time was for an encoder improvement and the 2nd time was for an incorrect setting. We hope the HDR section finishes encoding around 11/1. Then we will render and encode the SDR content, which has 1/6th the amount of patterns.

One of the things we figured out back in July was how much better we can improve compression quality by not using placebo mode. It turns out, for test patterns, placebo actually degrades the quality. It took some brute force to identify the settings causing the loss. To be clear, this is a test pattern issue, not an encoder issue. An encoder is not trying to lossless encode, in lossy mode. So we are now encoding every test pattern 96 times to find the best settings per pattern. Its a PITA! But worth the results.

Right now I am working on the resolution menu. There are three types of patterns: Multiburst, Wedge and Zone Plate. There are 18 versions of each. There are 6 nit levels of each version. This means 324 patterns in total. Encoded 96 times each is 31,104 encodes I am running right now! Overkill not spoken here! This started on Monday and as of today, I am 1/3rd through them. I guesstimate it will take around 9-days to finish the test encodes. Then it will take a couple of hours for the actual encode for the disc.

I had actually run through all of this back in July, but I only did one nit level per pattern. This time I am doing each nit level as it turns out the encoder settings can be different for the best results. To be clear, these are only patterns that are different at each nit level. For patterns that are the same, and only metadata changes, we only encode once. We wrote a tool that changes the metadata after the fact so we don't have to encode the same pattern 6 times.

We are almost done writing the pop-up help. We should finish the first pass of all help text this week. Then I will do a once over and make notes. Then we will address the notes. Then I will share the help text with others to get additional feedback. Then we have to figure out if these can work as before using menus or if we need to put the help in an encoded video to make it work.

After resolution, we have Contrast Ratio, ADL, Backlight Resolution, Motion Evaluation, Motion Evaluation HFR and all the window patterns to encode. The motion patterns will probably take 3+ weeks to encode. The rest are less than a day each given that they are all static. I also need to en-encode all of the audio graphics. The slowest part is the pre-encode or brute force encodes to determine the best settings.

We are currently waiting on panning audio graphics, DTS:X, montage music (New music and Atmos mix) and Atmos A/V Sync. I believe that is it.
 

·
Registered
Joined
·
5,580 Posts
I put together some TIFF images of the Seattle shot with the Space Needle at each stage of processing so you can see how it evolved from start to finish. The first six images are in the REDWideGamutRGB / Log3G10 space, so they will appear to lack saturation and contrast when viewed on a gamma display. Also note that white balancing did not occur until images 7 and 8, when they were graded.

https://www.dropbox.com/s/iemk0ukdil961b5/SM_Processing_Example.7z?dl=1

1. 00_Legacy.tif - This is the legacy demosaic algorithm.
2. 01_IPP2.tif - This is the IPP2 demosaic, which retains more detail than legacy.
3. 03_IPP2_TNR.tif - This is the IPP2 image with temporal noise reduction applied.
4. 04_IPP2_TNR_Decon_Mask_C00.tif - This mask is showing the pixels that deconvolution will touch, which is every pixel in this example.
5. 05_IPP2_TNR_Decon_Mask_C05.tif - This mask shows the settings I used for this shot when applying deconvolution. This way I avoid the noise in the sky getting sharper. Each shot used custom settings for best quality on the new disc. For the current disc, every shot used the same settings.
6. 06_IPP2_TNR_Decon.tif - This is IPP2 w/ temporal noise reduction and deconvolution.
7. 07_IPP2_TNR_Decon_Grade_HDR.tif - This is image number 6 graded for HDR (New disc) (BT.2020 / ST2084)
8. 08_IPP2_TNR_Decon_Grade_SDR.tif - This is image number 6 graded for SDR (New disc) (BT.709 / Gamma 2.4)

These are 16-bit full range TIFFs.

If you have Photoshop, I would load all images as a stack. Then you can turn on and off each layer to see the difference as you look at the same portion of the image.
 

·
Registered
Joined
·
1,694 Posts
Thank you for the update and I can't wait for the new disc to become available as I didn't have a need the first release on UHD.
 

·
Registered
Joined
·
5,580 Posts
@sspears Could you define backlight resolution?
In the most simple turns, you can think of it as a contrast ratio type of measurement. Ours involves 10 measurements. We will provide an xls on our website to download that you plug the numbers into and get a result back. We hope reviewers will find it useful as another metric to differentiate displays.
 

·
Registered
Joined
·
3,180 Posts
In the most simple turns, you can think of it as a contrast ratio type of measurement. Ours involves 10 measurements. We will provide an xls on our website to download that you plug the numbers into and get a result back. We hope reviewers will find it useful as another metric to differentiate displays.
I did a little reading, its a very interesting metric. I have a nagging feeling that one company has already, invested in its use.

Sent from my LGMP450 using Tapatalk
 
1221 - 1240 of 1254 Posts
Top