HDR Discussion Thread

[Geoff D]

Well played, sir.

[PaulGo]

Interesting article on UHDTV - HDR and WCG.

http://www.lightillusion.com/uhdtv.html

[Stacey Spears]

Originally Posted by Penton-Man

On the other thread I see you liked Yedlin’s stuff -

Having a 16bit file in an ACES workflow as a digital intermediate source throughout the post is important for an optimal and efficient process to picture quality outcome.

I would like to know what Yedlin used for his compression demo. Blu-ray does not have the quality loss that his compression demo had.

On the CMOS math, to have true UHD resolution in the red and blue channels, you need to shoot at QUHD.

On the subject of 16-bit, I recently got verbally attacked by someone who shoots commercials, claiming no one is using 16-bit files in post. I guess he had never heard of ACES or OpenEXR. My VFX friend could not stop ranting when I told him about this. I chose not to get into the debate, not worth it.

I remember the time I had asked Annie about ProRes. She looked at me and told me she did not work on TV. Made me laugh hard!

[Stacey Spears]

Originally Posted by Penton-Man

Steve actually presented much of his assertions that the selection of scaling algorithm has more impact on the perceptual experience than some people might have assumed

Have I shared any examples of the new down scaling algorithm I have been working on? I recently shared it with some friends at Pixelogic. Still need to do some final tuning on it using a QUHD and UHD display so that they perceptually match. They should be close now, but I want to verify it on the two displays for more empirical data, though it is subjective when looking at them side-by-side. Not to mention the number of images you need to look at vs. tuning on a single image.

It has been optimized for 2:1 downscale. (4320p to 2160p or 2160p to 1080p) It has a really nice anti-ringing filter. I had not realized just how much ringing was in down scaled images until I started working on this. For > 2:1 (e.g. 4320p to 1920p), need to use a different filter kernel to reduce moire. I know what will work, just need time to implement it.

Next up is a new chroma down scaling algorithm (based on the 2:1 down scaling algorithm mentioned above) to convert to 4:2:0. Here is a new chroma up scaling algorithm (4:2:0 to 4:4:4). I would be curious to know what people think. Bilinear was used to convert from RGB to 4:2:0 on all of them. Only thing different is the up scaling algorithm. The color bars has three examples while the other two have two examples. Need to do better cropping so they are easier to compare, but it should be good enough to look at. Once I get the new chroma down scaling working, then I can re-run the up scaling on these and see if they are further improved. While we can't control the up scale, and most devices appear use bilinear (in my testing), we can control the down scale. Maybe through a future SMPTE paper / presentation, we can get people to implement the chroma up scaling shown above.

[Geoff D]

Originally Posted by Stacey Spears

I would like to know what Yedlin used for his compression demo. Blu-ray does not have the quality loss that his compression demo had.

On the CMOS math, to have true UHD resolution in the red and blue channels, you need to shoot at QUHD.

If you look at his little presentation thingy in the right hand corner (that he clicks on to go to his next image) it says "netflix recommendations" or something similar so when he talks about compression and gives out that hideous example - which we know is nothing like what Blu-ray and/or UHD can do - he seems to be using bit-starved streaming bitrates as his guide which is a little odd.

As for Bayer arrays and whatnot then yes, shooting in higher resolutions will get you greater chroma resolution in the downscaled end product, so with red and blue only having a quarter each (and green the other half) you would indeed need a 16K Bayer array for true 4K red and blue, and a 8K array for true green in 4K. And yet I still love how the 6K Alexa 65 looks in his comparisons, it beats everything by quite a comfortable margin...or so it would appear. (I'm patiently waiting for the day that ARRI string together three of those A3X chips to create a true 15-perf 65mm sized sensor. )

You're a big proponent of RED, what say you about his image comparisons?

[DanBa]

Originally Posted by DanBa

"Dolby tone/gamut mapping."
What does it mean?

Does it mean Intelligent Display Mapping Engine’s algorithm (i.e. Dolby Vision Display Mapped Playback’s algorithm) + Dolby Vision dynamic metadata?
In other words, the Dolby Vision dynamic metadata specification can’t itself describe all the features of the (reference) Dolby tone/gamut mapping.

Originally Posted by Geoff D

...
Anyhoo, yes, DV's special sauce would be useless if every display just applied it willy nilly, so every DV display type needs its own 'golden reference' profile (usually averaged from several test models) loaded into the Dolby processing engine.

Thanks!

So, the proprietary Dolby Vision is a completely designed HDR format:

Proprietary Dolby Vision = ST 2084 PQ + ST 2086 static metadata + proprietary ST 2094 dynamic metadata + reference (proprietary) source implementation

For the time being, a HDR format based on the open standard ST 2094 (ST2094-10 or ST2094-40 for example) is an incomplete HDR format: a reference source implementation is missing.

[Penton-Man]

Originally Posted by Stacey Spears

On the subject of 16-bit, I recently got verbally attacked by someone who shoots commercials, claiming no one is using 16-bit files in post. I guess he had never heard of ACES or OpenEXR.

or sat in on some sessions at for instance Theater1 at Technicolor.

Originally Posted by Stacey Spears

My VFX friend could not stop ranting when I told him about this.

they appreciate it when you give them 16bit floating point Open EXR files with ACES AP0 primaries

[Penton-Man]

Originally Posted by Stacey Spears

I remember the time I had asked Annie

At one of those roundtables?

Under her creative theatrical watch/supervision, wrt a fairly recent feature film for the DV HDR finish, the sdr odt was replaced with a non proprietary ODT in order to optimize the images for the Dolby Vision projector, then subsequently a different ODT was utilized for the home HDR version by other staff.

[Penton-Man]

Originally Posted by Stacey Spears

...I would be curious to know what people think.

I need much more time. Perhaps later, we’ve got to get some of the old band back on here at Blu-ray.com. Does anyone know if madshi is still active?

If nothing else, he could help us Yankees translate YouTubes from German to English....

Sony PXW-FS5 und PXW-Z150 HDR Update und verbesserter FS5 Griff - NAB-Show Report 2017 - YouTube

[Penton-Man]

Originally Posted by Stacey Spears

color

a little history and a look at cinema engineers –

The Future of Cinema-Projection and Display Technology - YouTube

P.S.
advanced question to anyone – who do RIT Motion Picture Sciences folk sometimes cite (in admiration) in some of their published papers?

[Stacey Spears]

Originally Posted by Geoff D

you would indeed need a 16K Bayer array for true 4K red and blue, and a 8K array for true green in 4K.

You're a big proponent of RED, what say you about his image comparisons?

Using his math, it looks like 8K (QUHD) should cover red and blue for UHD delivery.

width * height = resolution

7680 * 4320 = 33177600
3840 * 2160 = 8294400

1/4 of them are red: 33177600 / 4 = 8294400
1/4 of them are blue: 33177600 / 4 = 8294400
1/2 of them are green: 33177600 / 2 = 16588800

I may have messed up my math, but it looks like 8K CMOS for 4K resolution and 16K CMOS for 8K resolution, which is what Astro does inside of their camera.

As far as Red, I wonder if he used the new or old debayer. The new debayer results in greater detail and less chroma artifacts. It is a bug improvement in image quality, in my opinion. So either the halation test could be better or worse, depending on the debayer used.

I was most surprised, not really, on how noisy the film was. I was a little disappointed with the 15-perf IMAX. At the same time, I have yet to see any digital camera produce the same image quality that I saw during the opening of the Dark Knight (bank robbery) in a real IMAX theater. That was so breathtaking!

While I am a fan of Red, I consider myself down to earth and know its strengths and weaknesses. I still remember the day that Claudio Miranda posted on Reduser and got ran off by some overzealous folks. I thought that was a depressing moment. Kind of reminded me of AVS forum and Don getting run off back in the day by some overzealous HD DVD folks.

I am one of the few that has the VistaVision Red. I received mine in December and have been shooting a doc on image quality and display calibration. Focus is on preserving artistic intent. I still have several to interviews to shoot, but I have interviewed Joe Kane, Michael Cioni, Ian Vertovec to name a few. Michael and Joe sure can talk. (A good thing) They are almost two hours each, which is close to 2 TBs of footage each.

I figure a doc on image quality should look good. At f/2.8, the DOF is so shallow. One eye is in focus and the other is not. I love the fall off. I am a fan of large format sensors. While VV is small compared to 65 and 15-perf IMAX, it is a nice jump up from S35. It looks like a full frame 35mm DSLR, in motion.

[Geoff D]

Yep, the depth of field being so shallow is one of those classic markers of large format content and iit'd be nice if VVLA could make a comeback, if only in the digital realm. RED have started things off and there's Sony's own "full frame" camera to come, as Penton first alerted us mere mortals to. They could even do a Tarantino and dust off the old 1.5x Technirama anamorphics to get a 2.25 widescreen image, that could then be laid down onto 70mm exhibition with little difficulty.

[Penton-Man]

Originally Posted by Penton-Man

P.S.
advanced question to anyone – who do RIT Motion Picture Sciences folk sometimes cite (in admiration) in some of their published papers?

Ans. ^ Prof Andrew Stockman of UCL IRIS and known to color scientists for http://www.cvrl.org/

How do we see colours? | Andrew Stockman | TEDxUAL - YouTube

[Stacey Spears]

Originally Posted by Penton-Man

Does anyone know if madshi is still active?

Mathias continues to improve madVR. He has a new neural network based up scaling that is pretty much better than anything on the market.

Our SSIM based based down scaling is based on the same paper. We have very different opinions on the primary filter kernel used. His is done in real-time with less precision (He has to do real-time playback) while we are offline and using double precision float. Slightly different anti-ringing algorithms. Here is a comparison of different madVR scaling algorithms. The SSIM / SSIM_AB is the specific algorithm. This clip needs AB (anti bloat) while most do not. AB mostly benefits up scaling. We don't use AB and are somewhere in between the two in terms of detail preserved. This is what we will tune with the QUHD and UHD displays side-by-side.

The chroma up scaling I showed was done in madVR. I had presented the idea to him and then he came back 6-months to a year later with the first version implemented. Shiandow did the original implementation and he and I have talked about presenting it in a paper. The idea is to use a cross-bilateral filter to use luma to recover the chroma when up scaling. Works well as long as you have a nice luma edge. When you don't, you need to fall back to something else. This is where more work needs to be done.

[Penton-Man]

Originally Posted by Stacey Spears

One eye is in focus and the other is not.

In real life, with some peoples’ eyes, it’s known as anisometropia

[Penton-Man]

Even though Geoff doesn’t have the opportunity to attend CineGear, he keeps up…. https://forum.blu-ray.com/showthread...e#post13734118

[Stacey Spears]

Originally Posted by Geoff D

They could even do a Tarantino and dust off the old 1.5x Technirama anamorphics to get a 2.25 widescreen image, that could then be laid down onto 70mm exhibition with little difficulty.

The sensor height is 21.6mm, which works pretty well for anamorphic.

[Penton-Man]

Originally Posted by Stacey Spears

I am a fan of large format sensors.

Even though Geoff doesn’t have the opportunity to attend CineGear, he keeps up…. https://forum.blu-ray.com/showthread...e#post13734118

[Stacey Spears]

Originally Posted by Penton-Man

a little history and a look at cinema engineers –

The Future of Cinema-Projection and Display Technology - YouTube


P.S.
advanced question to anyone – who do RIT Motion Picture Sciences folk sometimes cite (in admiration) in some of their published papers?

Thank you for sharing, had not seen that video. Nice to see David in the video.

[Geoff D]

Originally Posted by Stacey Spears

The sensor height is 21.6mm, which works pretty well for anamorphic.

If you mean regular 2x anamorphic then yeah, but I mean the format which was devised purely for VVLA, someone could resurrect it as Tarantino did for the old Ultra Panavision glass on Hateful 8. Heck, I'll bet that someone will have a crack at it in the next 5 years.