HDR heatmaps and torch mode UHDs

[Fjodor2000]

On my wishlist up for grabs by anyone curious to analyze:

[dorian]

Originally Posted by Fjodor2000

@dorian:

I noticed that you posted a HDR plot in the Heat 4K BD thread today:

https://forum.blu-ray.com/showthread...3#post21332443

To my understanding it was done using some new script that you have created. Is this correct, and if so have to added it to the analysis tools used by you and others in this thread?

I think it could be very useful for analyzing other movies too, so that would be much appreciated.

Also two questions:
* Since this script will analyze each pixel in each frame on the disc, approximately how long does it take to create a HDR plot using this new script? I know this will depend on PC hardware and length of movie, but would be nice to get a rough estimate (and maybe if you also can mention the CPU used, for reference).
* How does it differ from the HDR plots published by e.g. Macatouille? Are his plots e.g. based on metadata from the disc instead?

I don't know exactly how other HDR plots are measured. I wrote a C script for specific aspect ratios, measuring every pixel for every frame. The YCbCr frames are fed directly to the script and processed to output the MaxCLL and MaxFALL values for each frame to a CSV. I then put the CSV through a python script that generates the area chart with matplotlib.

The process is slow and impractical for most titles. If there are titles with contentious light levels or disputed metadata values, it can be useful.

Here are the MaxCLL & MaxFALL frames for heat:

[Show spoiler]

Here are some of the high MaxCLL frames:

[Show spoiler]

Here are some of the high FALL frames:

[Show spoiler]

[Macatouille]

The Lord of the Rings: The Fellowship of the Ring (2001)
Theatrical Cut

Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 602 cd/m2
Maximum Frame-Average Light Level        : 187 cd/m2

Heatmaps

[Show spoiler]

Gamut Visualizations

[Show spoiler]

HDR10 Plot

[Macatouille]

Given the length and popularity of the trilogy, I went a bit hog wild with the caps.

Not the most aggressive HDR grade in the world but I did laugh at Frodo in the tanning bed:



A couple stand-out uses of the WCG (there weren't a ton to choose from):

[dorian]

Babylon (2022)

I'm actually not a big fan of HDR but this is a good case of HDR graded purposefully. It's bright but doesn't feel unnaturally so. (The extremely high peaks measured are due to dust specks.)

MaxCLL:

[Show spoiler]

MaxFALL:

[Show spoiler]

The shots are representative of the MaxCLL & MaxFALL peaks, not the averages:

[Show spoiler]

[Show spoiler]

[Macatouille]

The Lord of the Rings: The Two Towers (2002)
Theatrical Cut

Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 805 cd/m2
Maximum Frame-Average Light Level        : 331 cd/m2

Heatmaps

[Show spoiler]

Gamut Visualizations

[Show spoiler]

HDR10 Plot

[Macatouille]

Gandalf takes a turn in the tanning bed this time:



Like FOTR, not a ton of stand-out WCG shots. Did like this one, though:

[Fjodor2000]

Originally Posted by dorian

I don't know exactly how other HDR plots are measured. I wrote a C script for specific aspect ratios, measuring every pixel for every frame. The YCbCr frames are fed directly to the script and processed to output the MaxCLL and MaxFALL values for each frame to a CSV. I then put the CSV through a python script that generates the area chart with matplotlib.

The process is slow and impractical for most titles. If there are titles with contentious light levels or disputed metadata values, it can be useful.

Here are the MaxCLL & MaxFALL frames for heat:

[Show spoiler]

Here are some of the high MaxCLL frames:

[Show spoiler]

Here are some of the high FALL frames:

[Show spoiler]

Cool! So any way that tooling could be shared?

Also, do you have any example execution time for how long a HDR plot took to generate for a specific movie using some certain CPU using that tooling? E.g. 1 hour, 4 hours, 12 hours, ... using e.g. AMD 7950X?

Also, did you detect any noticeable difference between the HDR plot created using that tooling vs HDR plot based on disc metadata (which I assume e.g. Macatouille is using)? I'm thinking your tooling may be more precise, since it is based on all actual frames from the disc. But it could also be that the difference varies per movie depending on how accurate the disc's HDR metadata is.

Also, come to think of it, for regular HDR10 - is there really any dynamic HDR metadata at all (i.e. metadata that varies throughout the movie)? I thought there was only static metadata. And dynamic metadata was for HDR10+ and Dolby Vision only.

[mrtickleuk]

Originally Posted by Fjodor2000

Cool! So any way that tooling could be shared?

I could be wrong, but I believe @dorian shared it on 17th June here:

Originally Posted by dorian

HDR Heatmap Colab Notebook

I've fallen off the development horse, so don't know what a "Colab Notebook" is :/

I did have a couple of ideas of my own, for what it's worth. On the plot of MaxCLL / MaxFall, it would be nice if the x axis showed both Frame numbers (20,000 / 40,000 / 60,000 etc) and elapsed times.

A bit like this, if i've got my sums right

---------------------------------------------------------------------
0          20,000       40,000     60,000      80,000       100,000
00:00      13:53        27:46      41:40       55:33        69:26

Using commas for the thousands to help readability. You could even use an interval which works out nicely for both measures, eg 18000 frames = exactly 12mins 30secs, so it'd give us 12:30, 25:00, 37:30, 50:00, etc.


The other idea I had was to include the timestamp of the brightest frame in the legend, ie

Maximum (MaxCLL: 1460,70 nits [frame 25,000/21:52], avg 258.31 nits)
Average (MaxFALL: 273.90 nits, avg 11.95 nits)
Minimum (max: 0.000 nits)

You could do the same for the darkest frame/timestamp I suppose too.

HTH

[jrcorwin3]

[dorian]

Originally Posted by mrtickleuk

I've fallen off the development horse, so don't know what a "Colab Notebook" is :/

A Colab notebook is just Google Colaboratory where you can share and run Python code over the cloud without needing a local interpreter or libraries installed.

I did have a couple of ideas of my own, for what it's worth. On the plot of MaxCLL / MaxFall, it would be nice if the x axis showed both Frame numbers (20,000 / 40,000 / 60,000 etc) and elapsed times.

Using commas for the thousands to help readability. You could even use an interval which works out nicely for both measures, eg 18000 frames = exactly 12mins 30secs, so it'd give us 12:30, 25:00, 37:30, 50:00, etc.

Like this? (I also added faint grid lines)

[Show spoiler]

The other idea I had was to include the timestamp of the brightest frame in the legend, ie

Maximum (MaxCLL: 1460,70 nits [frame 25,000/21:52], avg 258.31 nits)
Average (MaxFALL: 273.90 nits, avg 11.95 nits)
Minimum (max: 0.000 nits)

You could do the same for the darkest frame/timestamp I suppose too.

HTH

This sounds like a good idea but different films have different frame rates (24fps vs 23.976fps). This doesn't seem like a big deal but it adds up over time and invariably leads to inaccurate timestamps.

Originally Posted by Fjodor2000

Cool! So any way that tooling could be shared?

No simple way tbh. The C script has to be compiled and run locally. It's a mess because I couldn't find a reliable way to detect the aspect ratio so resorted to many different iterations. Babylon for example took 7 hours to write the CSV. It's very slow and only worth it for titles with contentious light levels, as I said.

[jrcorwin3]

Originally Posted by dorian

A Colab notebook is just Google Colaboratory where you can share and run Python code over the cloud without needing a local interpreter or libraries installed.



Like this? (I also added faint grid lines)

[Show spoiler]

This sounds like a good idea but different films have different frame rates (24fps vs 23.976fps). This doesn't seem like a big deal but it adds up over time and invariably leads to inaccurate timestamps.



No simple way tbh. The C script has to be compiled and run locally. It's a mess because I couldn't find a reliable way to detect the aspect ratio so resorted to many different iterations. Babylon for example took 7 hours to write the CSV. It's very slow and only worth it for titles with contentious light levels, as I said.

[Fjodor2000]

Originally Posted by dorian

No simple way tbh. The C script has to be compiled and run locally. It's a mess because I couldn't find a reliable way to detect the aspect ratio so resorted to many different iterations. Babylon for example took 7 hours to write the CSV. It's very slow and only worth it for titles with contentious light levels, as I said.

Isn't there any associated Makefile or similar that can be used to compile the C code? Optionally a pre-compiled binary executable could be shared for Windows or Linux.

Or perhaps the C code could be converted to Python to make it OS generic.

W.r.t. automatic aspect ratio detection, didn't you invent some logic for that when creating the tooling for HDR / WCG analysis of single frames? Could it be reused for this purpose?

[dorian]

Originally Posted by Fjodor2000

Isn't there any associated Makefile or similar that can be used to compile the C code? Optionally a pre-compiled binary executable could be shared for Windows or Linux.

Or perhaps the C code could be converted to Python to make it OS generic.

W.r.t. automatic aspect ratio detection, didn't you invent some logic for that when creating the tooling for HDR / WCG analysis of single frames? Could it be reused for this purpose?

The C code cannot be converted to Python as Python has no library to directly feed the YUV frames to the script with ffmpeg. As for aspect ratio, I tried that but it was unreliable so resorted to literally skipping a certain number of lines on top and bottom for different aspect ratios.

[Macatouille]

The Lord of the Rings: The Return of the King (2003)
Theatrical Cut

Mastering display luminance              : min: 0.0050 cd/m2, max: 4000 cd/m2
Maximum Content Light Level              : 705 cd/m2
Maximum Frame-Average Light Level        : 120 cd/m2

Heatmaps

[Show spoiler]

Gamut Visualizations

[Show spoiler]

HDR10 Plot


Past Posts

[Show spoiler]

• 12 Monkeys
• Ad Astra
• Akira
• Alien
• Almost Famous
• Apocalypse Now (Final Cut)
• Back to the Future
• Back to the Future Part II
• Back to the Future Part III
• Bad Boys
• Bad Boys 2
• Battle Royale
• Blade Runner: The Final Cut
• Dune
• Edge of Tomorrow
• Godzilla
• Hereditary
• Hocus Pocus
• Indiana Jones and the Temple of Doom
• Interstellar
• The Lord of the Rings: The Fellowship of the Ring
• The Lord of the Rings: The Two Towers
• Raiders of the Lost Ark
• Road House
• Thelma & Louise
• The Usual Suspects

[Macatouille]

Funny to me that most of these shots are within Rec.709, with a couple exceptions, and then....Mount Doom:

[jrcorwin3]

[jrcorwin3]

[jrcorwin3]

[jrcorwin3]