Dolby Vision now possible through MP4 Mux.

[Gatorman]

IMO, the L2 trims can only be done/edited by the colorist/director so for our Resolve generated RPU, I recommend removing or overwriting the automatic trims otherwise they can make bright content very dark and dark content brighter which might not be the intended look.

Is this still your preferred method with (3-1)?

[RESET_9999]

Is this still your preferred method with (3-1)?

Yes, the L2 trims are removed by default in (3-1). I watched a couple of CM_analyze generated movies and it looks better than ever.
The latest Dolby algo is much better than the old one that Resolve uses and the fact that we can use HRD10plus scene cuts(or any official ones) is great. example: https://slow.pics/c/3EyCFPu9

[thirdmoon]

Yes, the L2 trims are removed by default in (3-1). I watched a couple of CM_analyze generated movies and it looks better than ever.
The latest Dolby algo is much better than the old one that Resolve uses and the fact that we can use HRD10plus scene cuts(or any official ones) is great. example: https://slow.pics/c/3EyCFPu9

Is there any way to speed up the process of 3-1. For me it is literally 10x slower than DaVinci. (30min DaVinci, 7-8hour on 3-1)

[RESET_9999]

Is there any way to speed up the process of 3-1. For me it is literally 10x slower than DaVinci. (30min DaVinci, 7-8hour on 3-1)

The only way is to not touch your PC while CM is running. It takes about 1h and the speed goes from 0.050 to 0.400
If I work on my PC at the same time, it can take 6h indeed. speed goes from 0.050 to 2.5

[skull88]

Yes, the L2 trims are removed by default in (3-1). I watched a couple of CM_analyze generated movies and it looks better than ever.
The latest Dolby algo is much better than the old one that Resolve uses and the fact that we can use HRD10plus scene cuts(or any official ones) is great. example: https://slow.pics/c/3EyCFPu9

Is there any way to speed up the process of 3-1. For me it is literally 10x slower than DaVinci. (30min DaVinci, 7-8hour on 3-1)

The question is how much do you care about the difference between cm_analyze vs. Resolve algorithm output in the final image, or even the (sometimes flawed/lazy/inaccurate) retail DV metadata, which is how 99.9% of the world sees the films, lol. Have you A/B tested a sample or movie yet? I find CM is best at this point, but Resolve is a close 2nd and in a blind-test, you may even end up preferring one or the other for a particular film. RESET is rightly focused on accuracy and upholding the original intent of the colourist/filmmakers as much as possible, but we're still working with approximation and no full Dolby Vision license, keep in mind. As far as the question of speed, Dolby obviously expects people to have dedicated server or workstation for running their pro tools, so your call how much time/resources you want to dedicate to this entire process, especially if you don't get much enjoyment out of the technical details/testing/troubleshooting aspect.

In the end, patience is the biggest virtue you'll have with all this, no full automation is possible without sacrificing quality control and not understanding the tools or workflows can lead to errors/issues too. Cheers!

[jlpsvk]

can't agree more..

right now, only flaw is, that latest cm_analyze is buggy and on cuda it's slowing down rapidly. or in other words, starting to use less and less gpu power... watched it recently, started at 26% GPU usage (LAST FRAME: 0.05s), and in one third of the analysis, it was down to 12% (LAST FRAME: 0.6s).

GPU used was GTX 1660 SUPER

[jlpsvk]

regarding slowdown of cm_analyze... has anybody tried nVidia's Studio driver????

[RESET_9999]

but we're still working with approximation and no full Dolby Vision license, keep in mind.

The license is useless and only needed for manual L2/L8 trim passes. Even if I had a license I would still remove the trims.
The only difference with the original DV that has no trims is the blended shots and scene cuts.

[skull88]

but we're still working with approximation and no full Dolby Vision license, keep in mind.

The license is useless and only needed for manual L2/L8 trim passes. Even if I had a license I would still remove the trims.
The only difference with the original DV that has no trims is the blended shots and scene cuts.

Agreed, what I meant was that we're only able to do so much, and we're not trying to actually replace the proper colourist/grading work that is ideally completed (including additional trims etc.). Leaving the higher L2 trims besides 100-nits from Resolve or other automated analysis is actually changing the image significantly, which I agree we should not do.

And jlpsvk, it doesn't seem to be related to CUDA, but possibly certain proRes files or another bug to do with CPU scheduling or something like that, it also happens when CPU-only run is forced. Most movies run fast for me entire run, even when using my PC, as long as I leave CM window open/foreground. Some slow down no matter what I do.

[RESET_9999]

Most movies run fast for me entire run, even when using my PC, as long as I leave CM window open/foreground. Some slow down no matter what I do.

yeah Windows 10 seem more stable with CM. On my Win11 I just cant touch my PC. Even browsing the web on firefox will slow down the analysis.
I doubt the problem is the prores file since the same file works at full speed in resolve. The script use a very old version of ffmpeg though(because of the audio 5.0 bug) so It might worth testing with the latest ffmpeg.

[skull88]

Most movies run fast for me entire run, even when using my PC, as long as I leave CM window open/foreground. Some slow down no matter what I do.

yeah Windows 10 seem more stable with CM. On my Win11 I just cant touch my PC. Even browsing the web on firefox will slow down the analysis.
I doubt the problem is the prores file since the same file works at full speed in resolve. The script use a very old version of ffmpeg though(because of the audio 5.0 bug) so It might worth testing with the latest ffmpeg.

Ah, interesting. I think all the files I have done proRes conversion independently first in Staxrip using ffmpeg newer version have not slowed down in cm, so you could be onto something? I might do that now anyway and will report if I get another video that slows down reliably and can try to send sample to compare. Cheers.

[thirdmoon]

Source - Prometheus.2012.COMPLETE.UHD.BLURAY

Script Used - DoVi_Scripts_1.9.2.6.beta_MKV
Comparison - https://slow.pics/c/ls8wBmji

Image 1 = CM Analyze Latest - 7 hours (Left it overnight without interference)
Image 2 =1080p Cubic Scaling (DaVinci) - 30min
(Ryzen 5 3600, GTX 1660ti Win10)

I tested the 1080p Cubic scaling DV and CM analyze DV on my TV (Sony 65X80K), and I couldn't personally tell any difference for this particular movie.
Also, 3-1 clipped that 600nit highlight (between 10,000-20,000frame) to 400nit which isn't really helpful?

I will be testing with Expendables 2010 which has higher overall brightness. Will update.

Still not able to see the benefit of 3-1. Genuine doubt.

[RESET_9999]

Source - Prometheus.2012.COMPLETE.UHD.BLURAY

Script Used - DoVi_Scripts_1.9.2.6.beta_MKV
Comparison - https://slow.pics/c/ls8wBmji

Image 1 = CM Analyze Latest - 7 hours (Left it overnight without interference)
Image 2 =1080p Cubic Scaling (DaVinci) - 30min
(Ryzen 5 3600, GTX 1660ti Win10)

I tested the 1080p Cubic scaling DV and CM analyze DV on my TV (Sony 65X80K), and I couldn't personally tell any difference for this particular movie.
Also, 3-1 clipped that 600nit highlight (between 10,000-20,000frame) to 400nit which isn't really helpful?

I will be testing with Expendables 2010 which has higher overall brightness. Will update.

Still not able to see the benefit of 3-1. Genuine doubt.

CM_Analyse with L1 tuning is the latest version of the Dolby Algo which evolved a lot since Dolby was created. It went from fully dynamic with very aggressive/higher L1 values to lower & less dynamic metadata(10nits/1000nits floor). I wonder if they did that because of the complaints you often hear about DV being darker than HDR10.

With dark movies like Expendables and Prometheus, the difference is small. You'll notice a more significant difference with a brighter movie like in the example below.

Higher L1 means:
- more tone mapping, less clipping, darker image compared to HDR10
Lower L1 means:
-less tone mapping, *more clipping, brighter image closer to HDR10
* more clipping only if your TV is not capable of displaying the actual brightness. Basically, it just means that the image is displayed with less tone mapping, similar to static HDR10.

And here's a direct comparison using my x800m2 and an HDMI capture card:
PQ: https://slow.pics/c/vnEnoTiU
Plot: https://slow.pics/c/3QzuYren

the main benefit of CM:
- you can use official studios scene cuts.
- latest/best version of the Dolby algo
- Black metadata improved
- No need to use some strange resolve settings (1080p cubic)
- CLI tool. can be scripted/batch mode.

Not sure why it's still slow even when you don't touch your PC because I did 5 movies last night in batch mode with workflow (3-2) and some movie does indeed take more time but nothing close to your 7h
These time results include: Prores conversion, CM analyse, demuxing, Injection and mixing.



first movie took about 1h30 to process
2nd 1h20
3rd : took 3h
4th: took 2h
5th: took 1h20

[thirdmoon]

Thanks for the clarification @RESET_9999

I will test some brighter movies. Also will test the force CPU function, I think that might make a difference.

[RESET_9999]

it slow down with the CPU too and overall, much slower than using the GPU cuda