Spears & Munsil 4K UHD HDR Benchmark (Test Discs) May 15, 2023

[Stacey Spears]

Originally Posted by dcforsyth

Stacey Spears,

Any idea when the new disc will be available?

Thanks

If we can start authoring in late June, then we should be able to have something out in the fall.

[mrtickleuk]

Originally Posted by Stacey Spears

What content is your source frames from? I have never seen their smoothing do anything at all, even on the Quant pattern of ours. I just turn it off.

It was someone's post to AVS a while ago. I just keep lots of pictures that look interesting/useful. I think it's from The Revenant, but I could very easily be wrong.

On the Z9D, their smoothing can even smooth out most of the 8-bit simulated section. So it at least does something, though I have turned it off too.

Yes, this all stemmed from someone who wants to use the feature and was paranoid that any given firmware upgrade broke it. I reassured them that the smoothing still works, but it got me to thinking - how hard it is to PROVE that it's working. Hence the idea for a test pattern for it, instead of having to look at random screengrabs!

I use it on "low" on my OLED since it's impossible to separate it from MPEG Noise reduction.

I have not found an OLED that shows the flashing on the patterns we have created. I think the luma lag is interesting and need to test that on a recent Panasonic when I can get near one.

Cool

[Geoff D]

Originally Posted by Stacey Spears

Slight correction, it was FEL, not MEL. What I sent was with FEL using deblock and FEL w/o deblock. Not sure which is better since the FEL is not really an image but data. Same with the intra smooth feature of x265.

Several improvements have gone into the new edit along with changes. Some stuff in purple on the current montage that are now blue in the new as they should be. The eagles beak is really clipped on the current that is now fixed.

LA Skyline:
-There is a door with a blue light around it that turned purple
-The building on the right with the three levels of glowing windows was clipped, so you could barely make out the slits between the windows. Now you can clearly see them.
-Sharper overall

Seattle Skyline (Space Needle)
-Crane on right should be blue but is purple on current disc

Seattle Ferris Wheel
-Neon sign with wings should be three colors and is two on current disc.

Quantum Dot
-Far right vile is not the correct hue on current and is now a darker blue as it should be

Lots of stuff like the above all throughout. Two major things occurred:
1. Resolve's color management does not properly pull in all colors outside of the output gamut. These were clipped in the last one. This time we exported linear EXRs that contained the negative pixel values.
2. Transkoder had caused some hue shifts while repairing the clipped gamut. It now had negative values to use and they improved the algorithm.

In addition, the deconvolution algorithm was improved and we used a optimized setting for every shot to get the most out of it vs. using a fixed setting for the entire thing. Was a lot more work.

Ran dozens of encodes for the base layer. Some interesting discoveries along the way. Same for the FEL. So compression using x265 is better than the Sony encoder used on the current disc. Sony's encoder always came in under our target bitrate by several Mbps.

Right now I am rendering the graded vs. ungraded version. Should be done compositing the two this time tomorrow and then I can start the slow encode. This will let you see the hot pixels, dust blobs, VFX suns, etc... Hopefully this will give people an insight into what a colorist brings to the table.

Stacey, what's the cause of the purple shift? It's something that's bugged me about HDR for a long, long time, not that it can't do "true blue" because it can, but when I do my SDR to HDR comparisons I've noticed that what is a strong blue on the SDR has now become a distinct purple tone in many, many HDR discs.

[Geoff D]

Originally Posted by mrtickleuk

It was someone's post to AVS a while ago. I just keep lots of pictures that look interesting/useful. I think it's from The Revenant, but I could very easily be wrong.

Yes, this all stemmed from someone who wants to use the feature and was paranoid that any given firmware upgrade broke it. I reassured them that the smoothing still works, but it got me to thinking - how hard it is to PROVE that it's working. Hence the idea for a test pattern for it, instead of having to look at random screengrabs!

I didn't look at your attached pics before but holy shit that's some awful banding. Is someone actually getting that on their TV, it's not a mock-up or anything? Would dearly love to know exactly what film it is so I can test it on the ZD9.

[mrtickleuk]

Originally Posted by Geoff D

I didn't look at your attached pics before but holy shit that's some awful banding. Is someone actually getting that on their TV, it's not a mock-up or anything? Would dearly love to know exactly what film it is so I can test it on the ZD9.

I've found (with google reverse image search) the original forum post from whence they came. Unfortunately a lot of the matches are from when I re-used it in other posts!

here: http://webos-forums.ru/post67441.html#p67441

the EXIF data in the files I have says the images were taken on Fri 13th Apr 2018. That post was Sun 15th Apr 2018, so it checks out. Unfortunately as it's not AVSforum or AVforums it's not so simple to ask the owner what it was from but whatever it was, it's the kind of thing I'd like in a test pattern as it really makes it obvious what the TV is doing!

edit: It then gets used in a spanish forum here: https://www.avpasion.com/foro/t/lg-o...e-2#post-36973 - but someone has taken the original on/off images, and made new on/off images from that firmware where it wasn't working. So they knew where it was from, to do the new test images.

edit2: Curiouser and curiouser! A nice editorial feature about that firmware here https://hdtvpolska.com/nowe-oprogram...ktualizacje/2/ says (google translated from Polish):

Such a false comparison as on these "gifs", where you can already see a solid amount of dithering from the very fact of using this format, we probably have not seen in life in the editorial. We don't know where it comes from, but for us someone wants to deliberately plunge the producer. Just look at posterization in the sky, which is unreal and looks even not serious. Clear banding in the bright parts of the image does not appear in principle with SDR content, and here it looks as if at most 16 shades of color in the film were used. A kind of nonsense and manipulation. Based on our experience, we can add that it is not possible for even an algorithm to deal with this in this way, because all de-counturing filters process mainly small tonal transitions, not cakes filling 1/10 the width of the TV.
Źródło: https://hdtvpolska.com/nowe-oprogram...ktualizacje/2/

It may be that the original 2018 images were faked. But someone (maybe someone else?) produced the 4.10.31 animated gif and went to some effort to do so!

[Kris Deering]

Originally Posted by Geoff D

Stacey, what's the cause of the purple shift? It's something that's bugged me about HDR for a long, long time, not that it can't do "true blue" because it can, but when I do my SDR to HDR comparisons I've noticed that what is a strong blue on the SDR has now become a distinct purple tone in many, many HDR discs.

Obviously not Stacey but I think this has to do with the way the color shifts as you desaturate. Some colors can't be maintained at a lower luminance level so they shift. So you have to figure out a way to maintain the color intended even though they wouldn't have been the same with the lower luminance point.

[Stacey Spears]

Originally Posted by Geoff D

Stacey, what's the cause of the purple shift? It's something that's bugged me about HDR for a long, long time, not that it can't do "true blue" because it can, but when I do my SDR to HDR comparisons I've noticed that what is a strong blue on the SDR has now become a distinct purple tone in many, many HDR discs.

Its a hue shift that occurs with tone or gamut mapping. It may be in the content or in the displays processing. In my case, it was the gamut mapping to legalize colors on the content creation side.

[Geoff D]

Thanks gents. I don't discount it being a display issue, have you observed such a phenom on your ZD9 Stacey? I've profiled and calibrated mine and the blue is tracking just fine in the numbers and the actual HDR test patterns look blue, properly blue, too. But so many UHDs seem to shift what should be blue to purple.

Originally Posted by Kris Deering

Obviously not Stacey but I think this has to do with the way the color shifts as you desaturate. Some colors can't be maintained at a lower luminance level so they shift. So you have to figure out a way to maintain the color intended even though they wouldn't have been the same with the lower luminance point.

Yes, when I reduce the contrast on my TV to be able to resolve 4000 nits of highlight information then the colour saturation drops off noticeably, so I have to nudge the colour up to compensate. This purple shift isn't directly related to that though, I use this reduced contrast setting very rarely these days and most stuff is watched either in Dolby or in my main 1000-nit viewing mode which tracks the EOTF correctly (and resolves colour properly in calibration) and yet the issue persists.

If it is a TV problem then I'm not sure what I can do to correct it as the ZD9 doesn't have a CMS, just 2 and 10-point white balance. An external LUT box is out of the question, mainly because I want to be able to eat, pay bills etc over the coming months

[Kris Deering]

Originally Posted by Geoff D

Thanks gents. I don't discount it being a display issue, have you observed such a phenom on your ZD9 Stacey? I've profiled and calibrated mine and the blue is tracking just fine in the numbers and the actual HDR test patterns look blue, properly blue, too. But so many UHDs seem to shift what should be blue to purple.



Yes, when I reduce the contrast on my TV to be able to resolve 4000 nits of highlight information then the colour saturation drops off noticeably, so I have to nudge the colour up to compensate. This purple shift isn't directly related to that though, I use this reduced contrast setting very rarely these days and most stuff is watched either in Dolby or in my main 1000-nit viewing mode which tracks the EOTF correctly (and resolves colour properly in calibration) and yet the issue persists.

If it is a TV problem then I'm not sure what I can do to correct it as the ZD9 doesn't have a CMS, just 2 and 10-point white balance. An external LUT box is out of the question, mainly because I want to be able to eat, pay bills etc over the coming months

When you say your "1000 nit viewing mode", are you only selecting this with 1000 nit or less titles? Otherwise you may have issues with titles that exceed that nit range (as I'm sure you know).

[Geoff D]

Yes, I use that mode only with 1000-nit titles, mainly stuff from Fox, Disnee, Paramount and Universal. The TV can hit ~1800 nits peak in this same viewing mode but that's brightness only, as colour begins to clip well before then according to Stacey's excellent Color Clipping pattern. I refer to it as my '1000-nit mode' because it represents such grades in "1:1" fashion, as you called it before, as all colours extend to at least 1000 nits in this mode but the key ones don't go much further. So when watching stuff that extends much higher - and I'm referencing average brightness as well as peaks here - I have to resort to the Panasonic HDR Optimiser to bring the colour volume back. Either that or use DV if it's available, which is a godsend on something like The Meg. The reduced contrast '4000-nit mode' that I used previously has been all but retired.

[Stacey Spears]

Originally Posted by Geoff D

Thanks gents. I don't discount it being a display issue, have you observed such a phenom on your ZD9 Stacey? I've profiled and calibrated mine and the blue is tracking just fine in the numbers and the actual HDR test patterns look blue, properly blue, too. But so many UHDs seem to shift what should be blue to purple.

Both the Z9D and LG C9 hue shift blue to purple at really high nit levels with HDR10. In Dolby Vision, no hue shift, on the C9. Have not tried Dolby Vision on the Z9D yet. Usually occurs when the content goes above ~4000 nits. At least on the test pattern.

It is one of our new patterns. I thought I shared the DV UHD BD image with you that has the hue shift pattern on it. It is the pattern right after the color clipping. It has colors stacked (709 in 2020, P3 in 2020 and 2020).

On the montage, it was in the content. The software is used by many, so it may be what you are seeing. Though, it really was for really bright blues near the edge of the spectral locus, like LEDs.

[Geoff D]

Originally Posted by Stacey Spears

Both the Z9D and LG C9 hue shift blue to purple at really high nit levels with HDR10. In Dolby Vision, no hue shift, on the C9. Have not tried Dolby Vision on the Z9D yet. Usually occurs when the content goes above ~4000 nits. At least on the test pattern.

It is one of our new patterns. I thought I shared the DV UHD BD image with you that has the hue shift pattern on it. It is the pattern right after the color clipping. It has colors stacked (709 in 2020, P3 in 2020 and 2020).

On the montage, it was in the content. The software is used by many, so it may be what you are seeing. Though, it really was for really bright blues near the edge of the spectral locus, like LEDs.

The TV runs out of road when stuff gets that bright in any case, yeah, I'm looking at the hue shift pattern now in HDR10 using my '1000-nit mode' and right at the end, at 10k nits, the blue is indeed skewing purple. But in all the values before then it still looks blue, genuinely blue. With the Panasonic Optimiser turned on OR watching in DV then it's blue all the way to the end on the ZD9, natch.

BUT I'm seeing the purple shift in real world content well below that level of 10k luminance, like the various neon-lit locales in Atomic Blonde (MaxCLL 600 nits) which are blue on the SDR disc but more of a purple on the UHD in HDR10, though the Dolby Vision does indeed make it look a touch bluer. La La Land is another one (MaxCLL 1000 nits), I mean on that film the 'royal blue' was a distinct part of the whole marketing angle, it's similarly blue on the SDR BD and yet to my eyes it's got various purple-tinged moments on the HDR UHD. Take the shot of the Tesla coil at the Griffith Observatory @ 55m04s, it's blatantly purple on the UHD while the BD is blue as anything. Even if I force DV onto LLL or SDR convert it (using the exact same TV settings to view it as the regular BD) the purple shift remains, which to me indicates that it's part of the HDR mastering there rather than the TV?

[Stacey Spears]

Originally Posted by Geoff D

BUT I'm seeing the purple shift in real world content well below that level of 10k luminance, like the various neon-lit locales in Atomic Blonde (MaxCLL 600 nits) which are blue on the SDR disc but more of a purple on the UHD in HDR10, though the Dolby Vision does indeed make it look a touch bluer. La La Land is another one (MaxCLL 1000 nits), I mean on that film the 'royal blue' was a distinct part of the whole marketing angle, it's similarly blue on the SDR BD and yet to my eyes it's got various purple-tinged moments on the HDR UHD. Take the shot of the Tesla coil at the Griffith Observatory @ 55m04s, it's blatantly purple on the UHD while the BD is blue as anything. Even if I force DV onto LLL or SDR convert it (using the exact same TV settings to view it as the regular BD) the purple shift remains, which to me indicates that it's part of the HDR mastering there rather than the TV?

That is probably the same issue I had where the software mastering the content introduced the hue shift. Assuming the SDR was a different grade and not made from the HDR, then that further points to the software used.

I had used a build of the software from last January (2019) for the current disc. The hue shift was greatly improved in a build from December. (2019) It was now blue, but different. I shared some frames with them and the beta build I have from early March (2020) has no shift at all.

[Stacey Spears]

Here are two sample crops.

1. Sorry they don't perfectly match.
2. They are in PQ, so the colors and look are not going to look correct on a computer monitor.

Current Disc


New Grade


Take a look at the door on the lower right. It is correct in the new grade.

The lower left should be purple.

Also look at the detail different in the top of the building. Detail was starting to clip in the old grade.

[Geoff D]

Originally Posted by Stacey Spears

That is probably the same issue I had where the software mastering the content introduced the hue shift. Assuming the SDR was a different grade and not made from the HDR, then that further points to the software used.

I had used a build of the software from last January (2019) for the current disc. The hue shift was greatly improved in a build from December. (2019) It was now blue, but different. I shared some frames with them and the beta build I have from early March (2020) has no shift at all.

Thanks Stacey, it's nice to know I'm not entirely insane. Though I'm amazed that something like this crept through QC (not at your end as you obviously caught it, I mean the makers of the mastering software itself) and that it's only just been corrected four full years into the format's life. I'm not however surprised that consumers never noticed because so many of them aren't running the correct colour temperature anyway, most preferring a 'cooler' look that naturally calms the purple down into a bluer colour.

[Stacey Spears]

Originally Posted by Geoff D

Thanks Stacey, it's nice to know I'm not entirely insane. Though I'm amazed that something like this crept through QC (not at your end as you obviously caught it, I mean the makers of the mastering software itself) and that it's only just been corrected four full years into the format's life. I'm not however surprised that consumers never noticed because so many of them aren't running the correct colour temperature anyway, most preferring a 'cooler' look that naturally calms the purple down into a bluer colour.

I thought the purple in my content was caused by something else. It was not until the re-grade that I realized it was something else.

Only the content creator will know if its wrong or not. As a consumer, we don't know what the colors are. I think the SDR vs. HDR examples that you provided are great! Of course HDR can produce colors that SDR cannot. Often bright blue goes white in SDR due to de-saturation to maintain luminance while staying blue in HDR since it can go much brighter before de-saturation.

What is funny about purple is this particular book. Seems appropriate for the current discussion.

[Geoff D]

Originally Posted by Stacey Spears

I thought the purple in my content was caused by something else. It was not until the re-grade that I realized it was something else.

Only the content creator will know if its wrong or not. As a consumer, we don't know what the colors are. I think the SDR vs. HDR examples that you provided are great! Of course HDR can produce colors that SDR cannot. Often bright blue goes white in SDR due to de-saturation to maintain luminance while staying blue in HDR since it can go much brighter before de-saturation.

What is funny about purple is this particular book. Seems appropriate for the current discussion.

The bolded did always come into my thinking whenever I encountered this issue, that it was just 'HDR doing what SDR cannot' and that the prevalent green tint of many an SDR grade was perhaps skewing those purples into into a more bluey-cyan kinda colour on the regular Blu. It's when you get something very modern like Atomic Blonde or LLL which have that 'proper' blue colour heavily represented even in their marketing which made me question what I was seeing on their respective UHDs. And they don't go especially bright, Blonde in particular, so it doesn't seem like a question of desaturating to maintain luminance.

I heard it in a HDR tutorial somewhere (it was a video of African wildlife I think?) that green was often added to SDR to make things like bright blue skies have more saturation, green being of a brighter 'wavelength' (or whatever the term is ) of light than blue so the skies can be pushed brighter without desaturating, as such it's uncanny how many UHDs I've seen that lose the pervasive greeny/yellowy tint from their SDR incarnations and have a more neutral colour balance. I wish it didn't come at the expense of the blue being skewed to purple (though it's not on every single UHD) but hopefully this is something that will be weeded out as the mastering software improves.

[Penton-Man]

Originally Posted by Geoff D

I heard it in a HDR tutorial somewhere (it was a video of African wildlife I think?)

Snippets taken from a 50 min-ish documentary in a presentation at the Dolby screening room in Soho by President Kevin Shaw (CSI - https://coloristsociety.com/ ) for an HDR workshop for the Guild of Television Cameramen, see this post for the link - https://forum.blu-ray.com/showthread...y#post15631804

[Stacey Spears]

Originally Posted by Penton-Man

Snippets taken from a 50 min-ish documentary in a presentation at the Dolby screening room in Soho by President Kevin Shaw (CSI - https://coloristsociety.com/ ) for an HDR workshop for the Guild of Television Cameramen, see this post for the link - https://forum.blu-ray.com/showthread...y#post15631804

I had not seen that before, thank you for sharing.

[Penton-Man]

Originally Posted by Stacey Spears

I had not seen that before, thank you for sharing.

You’re welcome . With the onset of the shelter-in-place guidelines, there have been more recent HDR-related free webinars/presentations posted online, check out the last several pages of the HDR Discussion thread.