Dolby Vision now possible through MP4 Mux.

[RESET_9999]

@RESET_9999 how did you generate the profile 7 RPU for Titanic?

With dovi_scripts Level(s) transfer workflow (2-3).
You must use the original scene cuts.

[staknhalo]

I dunno what the hell is going on with these titles (besides whoever not caring)- but if you guys look at the web and disc DV plot for Trap 2024 and web DV plot for Longlegs 2024 - there's almost no point to the DV metadata it looks like - yet the HDR10+ (from web for each) shows a lot more 'activity' - I only really looked and noticed it with these two so far, wonder how many others (I'm guessing since they started doing DV and HDR10+ together on recent iT/MA webs) are like this

Here's Trap to show what I mean

L1_plot.png (93.65 KiB) Viewed 25952 times

hdr10plus_plot.png (123.03 KiB) Viewed 25952 times

[RESET_9999]

It's because DV has a 100nits floor(except the earliest version of the algo) for L1 and values under 100nits are not allowed.
It wouldn't matter anyway because any L1 value below your TV target tells the TV to not do any tone mapping. So if you have a 800nits TV, any value between 100-800nits wont have any effect(as long as avg is not too high).

For example on my 800nits C2:

avg pq = 10nits:
rpu starts dimming around 850-900nits L1, any metadata below that has no effect (image is clipped because of that) and the higher the metadata gets, the dimmer the image gets.

avg_pq = 92 nits:
rpu start dimming earlier, around 650-700nits L1

avg_pq higher:
rpu start dimming even earlier.

You can test your own TV target relation with L1 using these files: https://drive.google.com/drive/u/1/fold ... HnkomkdsXm

P-S I re-graded both movies

[staknhalo]

It's because DV has a 100nits floor(except the earliest version of the algo) for L1 and values under 100nits are not allowed.
It wouldn't matter anyway because any L1 value below your TV target tells the TV to not do any tone mapping. So if you have a 800nits TV, any value between 100-800nits wont have any effect(as long as avg is not too high).

For example on my 800nits C2:

avg pq = 10nits:
rpu starts dimming around 850-900nits L1, any metadata below that has no effect (image is clipped because of that) and the higher the metadata gets, the dimmer the image gets.

avg_pq = 92 nits:
rpu start dimming earlier, around 650-700nits L1

avg_pq higher:
rpu start dimming even earlier.

You can test your own TV target relation with L1 using these files: https://drive.google.com/drive/u/1/fold ... HnkomkdsXm

P-S I re-graded both movies

So for titles like that would I be ok throwing DV out and just going HDR10 if not regrading/using your DV, or I should keep it for the color/trim/other stuff in the other levels of metadata still, right (because there is still some actual data there in L1 showing it's not empty, even if just being ignored/discarded as it should as you said)?

rtaImage.png (498 KiB) Viewed 25937 times

Was this maybe what was going on with the French Dredd 2012 disc too or was that one still just empty/blank metadata?

[RESET_9999]

It depends on how your TV handles the HDR10 static metadata and how the HDR10 encode metadata is set.
For example if the movie HDR10 doesn't have maxcll/fall avlues with a 4000nits MDL, the tone mapping curve your TV will use might introduce some tone mapping even if the content is at SDR level.

in other words, it's safer to use DV even if the values are within your TV target.

[staknhalo]

It depends on how your TV handles the HDR10 static metadata and how the HDR10 encode metadata is set.
For example if the movie HDR10 doesn't have maxcll/fall avlues with a 4000nits MDL, the tone mapping curve your TV will use might introduce some tone mapping even if the content is at SDR level.

in other words, it's safer to use DV even if the values are within your TV target.

So, basically for sake of argument even if there's only CLL and FALL data in a DV container 'properly' graded 0-2k/4k/whatever with literally no other metadata at all in the DV container on any level, it's still better than 0-1k graded HDR10 - because then as long as the TV has DV - you don't have to worry how the HDR10 of the TV is - that's the jist of it right?

I'll have to take a look at that Dredd disc again - I went with USA HDR10 only disc cause I think the L1 was blank when I looked at it on the French one - I think Longlegs I threw the DV out too

Edit - it's also so dumb they built 'clipping off' into HDR10 spec as like 'yeah just have TVs blow shit out' - as they also try to sell you on 'filmaker mode' or whatever

[RESET_9999]

No,
static DV metadata for a movie graded in the 1000-4000nits is not good and probably worse than static HDR10 because DV metadata is a lot more aggressive than HDR10 static curves.
For example if you have a 4000nits grade with a static 1000nits DV metadata. All the low-nits scenes will be darker and the brighter scenes will be more clipped than proper HDR10 metadata tone mapping curve. It's practically the same as making a DV P8 hybrid using two sources with different brightness, as shown here: https://www.youtube.com/watch?v=236athosits

Those movies with static DV are not done properly and should be avoided...
Trap is done properly because the content is below the DV 100nits floor (min)

[staknhalo]

No,
static DV metadata for a movie graded in the 1000-4000nits is not good and probably worse than static HDR10 because DV metadata is a lot more aggressive than HDR10 static curves.
For example if you have a 4000nits grade with a static 1000nits DV metadata. All the low-nits scenes will be darker and the brighter scenes will be more clipped than proper HDR10 metadata tone mapping curve. It's practically the same as making a DV P8 hybrid using two sources with different brightness, as shown here: https://www.youtube.com/watch?v=236athosits

Those movies with static DV are not done properly and should be avoided...
Trap is done properly because the content is below the DV 100nits floor (min)

got it - so essentially as long as there's even just 'a little' data/grading in the L1 (as in native Trap and Longlegs) it's 'good' DV data - French Dredd still bad - thanks

Edit - I verify pre-aligned remuxes with the web source or different disc release just by quickly glancing at L1 which is why I wanted to be sure instead of aligning the web data to disc myself to be lazier lol

[LV8HD]

@RESET_9999 in the most recent update you introduced the possibility to measure brightness of an HDR10 video on a frame-by-frame basis, which, to me, seems the most correct approach. Does this mean the MadVR method is less accurate even with excluded letterboxes? does it work on a scene-by-scene basis?

[RESET_9999]

@RESET_9999 in the most recent update you introduced the possibility to measure brightness of an HDR10 video on a frame-by-frame basis, which, to me, seems the most correct approach. Does this mean the MadVR method is less accurate even with excluded letterboxes? does it work on a scene-by-scene basis?

good catch haha. I didn't mention it in the release note because it's soooooo slow that it's unusable. I practically wasted my time on this code but it is indeed a frame by frame analysis / plotting. It uses the same script that generates the maxcll/fall stats in (7-2) (7-1).

Madvr analysis is also frame by frame but because I use Dovi_tool to plot the data, the graph is shot by shot. It's still 100% accurate but the values are rounded shot by shot. MadVR overall should be more accurate too since it measure pixels perceptually while the other script just measures the pixels as they are without any restriction.

[kazuma]

reset_9999, how did you find the PQ of the tenant with cm4.0?

[RESET_9999]

I dont know, never watched it. I dont like old movies.

[LV8HD]

@RESET_9999 in the most recent update you introduced the possibility to measure brightness of an HDR10 video on a frame-by-frame basis, which, to me, seems the most correct approach. Does this mean the MadVR method is less accurate even with excluded letterboxes? does it work on a scene-by-scene basis?

good catch haha. I didn't mention it in the release note because it's soooooo slow that it's unusable. I practically wasted my time on this code but it is indeed a frame by frame analysis / plotting. It uses the same script that generates the maxcll/fall stats in (7-2) (7-1).

Madvr analysis is also frame by frame but because I use Dovi_tool to plot the data, the graph is shot by shot. It's still 100% accurate but the values are rounded shot by shot. MadVR overall should be more accurate too since it measure pixels perceptually while the other script just measures the pixels as they are without any restriction.

I see. Correct me if I'm wrong. MadVR measures pixels using perceptual quantization while the other script the raw data?

[RESET_9999]

right

[LV8HD]

@RESET_9999 pushed by curiosity I decided to check what the fastest method of measuring HDR with excluded letterboxes was on my PC (MKV measurement vs. Cropped ProRes measurement). The winner ended up being MKV although I have a good CPU (AMD Ryzen 7 7700x), but that's not the point.

As soon as both measurements were done (same movie) I took a look at them expecting to see the same results, however, they differed of a couple of nits. Why is that?

MKV measurement


ProRes measurement