Cineform or DNxHD intermediates for editing in Resolve 14

Get answers to your questions about color grading, editing and finishing with DaVinci Resolve.
  • Author
  • Message
Offline
User avatar

cevisuals

  • Posts: 7
  • Joined: Mon Apr 30, 2018 4:47 pm
  • Location: Eugene, Oregon
  • Real Name: Cliff Etzel

Cineform or DNxHD intermediates for editing in Resolve 14

PostSat May 12, 2018 6:22 pm

New Resolve user having finally kicked Premiere Pro and Vegas Pro to the curb.

I've ALWAYS converted my highly compressed h264 clips to Cineform due to the 32bit nature of DNxHD clips having to be wrapped in a .MOV wrapper in order for Premiere CS6 or Vegas Pro to read them. So I've always converted to Cineform AVI's before I would start editing to take the load off the CPU - it works pretty well in Premiere, but Vegas is another matter (Decode error messages within Vegas would cause a hard crash)

Back to my question: Should I convert my native clips to an intermediate for editing and grading in Resolve? I know that 12.5 did not like h264 clips on the timeline and would never play back well for me.

I always have used Cineform but I'm also aware DNx is as relevant in the post world. Does Resolve prefer one over the other or is it a non-issue?
Last edited by cevisuals on Mon May 14, 2018 3:24 pm, edited 1 time in total.
Cliff Etzel
Photographer, Storyteller, Humanitarian
www.cliffetzel.com
Offline
User avatar

waltervolpatto

  • Posts: 5906
  • Joined: Thu Feb 07, 2013 5:07 pm
  • Location: Burbank, CA

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSat May 12, 2018 9:02 pm

Dnx is the way to go. Wildly used in post: higher compression for editing and lower for finishing
SuperServer 5039AD-I
C9X299-PGF - DDR4-2400 16x4 GB
i9-7920xCPU 12c 2.90GHz Water cooled
2x NVIDIA GeoForce GTX 1080ti
DeckLink Studio 4K
Window10 - Resolve Studio

Inb4YouCanJustColorComments: yes you can, good luck with that.
Offline
User avatar

Micha Clazing

  • Posts: 126
  • Joined: Sat Apr 01, 2017 3:26 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSat May 12, 2018 10:37 pm

Advocate of the devil here: h.264 absolutely flies in Resolve 15, on both nvidia and AMD. I can even play clips backwards with JKL, normally that would start chugging instantly because you're "going against the grain" of the frame delta compression, so to say.

But yeah, like Walter says, DNxHR LB is great for 4K editing (SQ for HD). 444 for VFX/keying, and HQ for grading/finishing. Only thing I would caution is Avid codecs have bitten me before with regards to video levels and the Adobe suite. Depending on the version of Premiere/After Effects, they'll either decode correctly, or you need to export as data levels in order to prevent a washed out image. I thought it was fixed since CC 2014 but I recently ran into the problem again so it must've regressed at some point.
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSat May 12, 2018 11:14 pm

I still have observed the same problem last year.
Can’t check if it’s gone now, since I stopped paying the rent ;-)

I’d be careful to recommend working in native H.264, it depends very much on your hardware.
But DNxHD/HR is fine, in DR15 it can even carry proper alpha.
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM
Offline
User avatar

Cary Knoop

  • Posts: 500
  • Joined: Sun Mar 12, 2017 6:35 pm
  • Location: Newark, CA USA

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 1:43 am

cevisuals wrote:Back to my question: Should I convert my native clips to an intermediate for editing and grading in Resolve?

Depends on the size, for HD I would not bother unless you have a really slow computer for 4K or higher I would always transcode unless it is a simple edit.

But make sure if your source has super-whites they do not get clipped by the intermediate codec.
Offline

Janis Lionel

  • Posts: 117
  • Joined: Mon Apr 04, 2016 8:09 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 10:23 am

Micha Clazing wrote:Advocate of the devil here: h.264 absolutely flies in Resolve 15, on both nvidia and AMD. I can even play clips backwards with JKL, normally that would start chugging instantly because you're "going against the grain" of the frame delta compression, so to say.


Odd...I really have struggles with H264 since Resolve beta 15! what system doyou use?
--
Windows 10 / i4930k @4.3Ghz / 32GB RAM / GTX 1080 / 12TB RAID 0 (4 x 3 TB) / Mini Monitor 4K / Eizo CG247x / Mainly working with CDNG 4:1
Offline
User avatar

Micha Clazing

  • Posts: 126
  • Joined: Sat Apr 01, 2017 3:26 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 10:37 am

Janis Lionel wrote:Odd...I really have struggles with H264 since Resolve beta 15! what system doyou use?

I work on a number of different systems, the highest-end of which has an nvidia 1080 in it. The AMD system I mentioned has a 390X in it and my laptop has a 940M. I always run Studio though, that might make a difference, especially on nvidia. Oh and Windows, never macOS.
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 10:59 am

No problems here on a mid-level iMac with Resolve Studio and H.264 in HD.
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 11:24 am

Janis Lionel wrote:
Micha Clazing wrote:Advocate of the devil here: h.264 absolutely flies in Resolve 15, on both nvidia and AMD. I can even play clips backwards with JKL, normally that would start chugging instantly because you're "going against the grain" of the frame delta compression, so to say.


Odd...I really have struggles with H264 since Resolve beta 15! what system doyou use?


It only works well if you deal with 4:2:0 8bit h264.
Anything 4:2:2 or 10bit means CPU decoding, so then you need very powerful machine.

h264 can have many variants, so be aware of it specially when you work with GH5.
You also want Resolve Studio license.
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun May 13, 2018 5:21 pm

I prefer to use cineform instead dnxhd/hr. I think avid codec have too much variance and not ever is clear if is 8/10/12 or upper bit. Also for alpha support (for graphics and more) i prefer cineform.
I use it from over 10 years with commercial version and i like its features.
Too much times i fight with editors to try to teach them the difference of bit for color between different dnxhd/hr incarnation.

Inviato dal mio E6653 utilizzando Tapatalk
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 12:20 am

You you get alpha in Cineform with DR?
I found only now with DR15 I can add alpha to DNx with MXF container, and that's what my collaborators want on the PC now.
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 7:02 am

I not tested alpha on Dr but i use a workflow where i have different softwares: compositing, 3D, editing and resolve for color then i prefer to use an only codec for all, good for win and mac, Adobe, blackmagic, and not only.
Uli Plank wrote:You you get alpha in Cineform with DR?
I found only now with DR15 I can add alpha to DNx with MXF container, and that's what my collaborators want on the PC now.


Inviato dal mio E6653 utilizzando Tapatalk
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 7:20 am

That's why I ask. I'm very much in favour of Cineform, but if I can't move alpha channels with it, it's not for me.
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 9:45 am

Cineform 16bit RGB option should preserve alpha.
Cineform with Alpha works well in Adobe apps.
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 10:47 am

I can confirm that it works with alpha now in DR15. Thanks for pointing this out, Andrew!

The only problem remaining is that it's only available in QuickTime. No problem for me, but some people got paranoid about using .mov files at all on the PC (completely misunderstanding the problem with QT framework).
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 10:50 am

If it comes to Adobe you don't need QT to be installed anymore, definitely not if you want to use Cineform. Latest AME is not using QT engine anymore as far as I understand. We've lost support for custom codecs though.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon May 14, 2018 12:04 pm

Uli Plank wrote:The only problem remaining is that it's only available in QuickTime [containers].


AFIK cineform support is only available in passive decoding mode but without encoding capabilities in ffmpeg and most other free software until now, whereas dnxhd and prores are supported in both directions and are IMHO much more compatible and suitable for real world exchange.
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 08, 2018 9:30 pm

Martin cineform is open source and you can implement encode decode free (with exception of raw that require a 20 dollar license), prores could be only decoded, if you want to encode prores officially you need to request and buy rights from apple. The reason be cause under windows few software can export prores officially. Ffmpeg encode prores not legally be cause they not have bought right to do it.


Inviato dal mio iPhone utilizzando Tapatalk
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon Jul 09, 2018 2:26 am

carlomacchiavello wrote:Martin cineform is open source and you can implement encode decode free (with exception of raw that require a 20 dollar license), prores could be only decoded, if you want to encode prores officially you need to request and buy rights from apple. The reason be cause under windows few software can export prores officially. Ffmpeg encode prores not legally be cause they not have bought right to do it.


well -- the actual license of cineform looks indeed much more acceptable, than in those days, when we all saw it as real attractive alternative. but in the meanwhile it isn't anymore a very hot topic. nevertheless there are still some significant cross platform/architecture incompatibles present, which make it a rather questionable choice for serious work resp. reliable exchange in practice -- see: https://trac.ffmpeg.org/ticket/6863

and i wouldn't worry to much about apple and legal actions against ffmpeg or any other "free" software, which uses this very well written alternative implementations of prores. that's at least on linux systems an acceptable situation, which works really satisfying in real world too.
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostTue Jul 10, 2018 11:35 am

Mmm what architecture no x86 they think to use for encode video? Arm? Or less performance arc?
I understand that ffmpeg must work on many different architecture, but if they not be able to encode correctly is better to adopt standard and limit the architecture then limit encoding capability...


Inviato dal mio iPhone utilizzando Tapatalk
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 1:05 am

carlomacchiavello wrote:Mmm what architecture no x86 they think to use for encode video? Arm? Or less performance arc?
I understand that ffmpeg must work on many different architecture, but if they not be able to encode correctly is better to adopt standard and limit the architecture then limit encoding capability...


i can not speak about the issues in this specific case, but architecture specific code/compilation optimizations are a really annoying topic. it's not only affecting different families of processors, but also very subtle differences within the feature sets of otherwise rather compatible CPUs. resolve users are well aware of this kind of troubles, because avids DNxHD/HR codec library, as it is utilized in resolve, is another very well example for this kind of issues. it makes use of some advanced SIMD operations, which are not available on AMD processors:

viewtopic.php?f=21&t=67554#p389563
viewtopic.php?f=21&t=68009

issues, which you will not face, if you use ffmpegs implementation of this format.

codecs can benefit a lot from more advanced CPU capabilities and optimized compilation. you just have to look at the really impressive encoding/decoding benchmark results of intels clear linux and compare it with most other linux distributions (e.g. the LAME and Zstd measurments in this recent comparison). but this kind of optimizations do not come for free. they always impair compatibility to some degree.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 9:15 am

Ffmpeg has its own issues. Lot of code which is not optimised and with 'shortcuts'. It still can't properly decode 12bit ProRes444 and Cineform RGB. It decodes them at 10bit only. This is quite a big flaw. In case of Cineform ffmpeg decoder is few times slower than reference one.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 4:44 pm

Andrew Kolakowski wrote:Ffmpeg has its own issues.


yes -- i agree! it doesn't make sense to glorify it without critical reservations.

Andrew Kolakowski wrote:Lot of code which is not optimised and with 'shortcuts'. It still can't properly decode 12bit ProRes444 and Cineform RGB. It decodes them at 10bit only. This is quite a big flaw.


the 8/10bit limitations in ffmpeg and some other important free video processing frameworks (e.g. MLT) is indeed one of the most annoying and limiting issues, if you want to utilize this open source solutions for serious work. i therefore try to use them only for common i/o-tasks, where the supported bitdepth is usually enough in most practical cases, but i definitely prefer more accurate data handling (e.g. by natron) for the actual image processing.

Andrew Kolakowski wrote:In case of Cineform ffmpeg decoder is few times slower than reference one.


well -- most of the hype around cineform is related to this optimized performance, otherwise you will will get just the same typical benefits and compression characteristics by any other wavelet based format (e.g. JPEG2K) as well. i really understand, that some people like this particular advantage, but it was always limited to a very small set of supported operating systems and hardware preconditions. sure -- this situation has changed a little bit to the better over the last year, but that came rather late -- too late(!), as i would say.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 5:30 pm

Cineform efficiency is at around ProRes and DNxHR level (bit lower), but easiness of decoding at fractional resolution is huge (underestimated) benefit of wavelet nature (also its artefacts nature is better suited for high quality masters). Jpeg2000 is complex and slow (=expensive to implement), so it never really gain real momentum. Cineform solves about all of these problems. It's plain simple, which is very good thing.

I'm pretty sure everyone would like ability to take 4K (or higher res) file out of BM camera and edit it straight away on simple laptop. This is what Cineform offers and it works perfectly in Premiere. No transcoding, no proxies- all is just a press of a button to jump from full res to half or quarter.

I've tested 3x4.6K Cineform files in Premiere at 50% opacity - switch to 1/4th resolution and you can edit it on i5 dual core :)
I don't understand why not a single company use ability of fractional decoding in case of ProRes (except Apple in FCPX). I have 80% valid confirmation that this is not Apple exclusive feature and it's present in reference SDK. It may be bit more complicated than in case of Cineform, but if it's in reference ProRes SDK then I simply don't get why companies don't use it. I'm also surprised there is no such a thing for DNxHR.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 5:31 pm

Martin Schitter wrote:.

Andrew Kolakowski wrote:Lot of code which is not optimised and with 'shortcuts'. It still can't properly decode 12bit ProRes444 and Cineform RGB. It decodes them at 10bit only. This is quite a big flaw.


the 8/10bit limitations in ffmpeg and some other important free video processing frameworks (e.g. MLT) is indeed one of the most annoying and limiting issues, if you want to utilize this open source solutions for serious work. i therefore try to use them only for common i/o-tasks, where the supported bitdepth is usually enough in most practical cases, but i definitely prefer more accurate data handling (e.g. by natron) for the actual image processing.


In this case it's big limitation which is against reference decoders. It's simply produces "wrong" decoded output in case of 12bit sources. This is without any warning etc. I don't understand why Cineform decoding is done to 12bit pixel format, but actual data is "internally" converted to 10bit only.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 6:41 pm

Andrew Kolakowski wrote:In this case it's big limitation which is against reference decoders. It's simply produces "wrong" decoded output in case of 12bit sources. This is without any warning etc.


please open a bug report at the ffmpeg issue tracker concerning this flaws!
that's the way how this kind of annoying deficits are usually handled and fixed in the open source world. :)

Andrew Kolakowski wrote:I don't understand why Cineform decoding is done to 12bit pixel format, but actual data is "internally" converted to 10bit only.


well -- i usually don't need 12bit codecs except for RAW sensor data in practice, and in those rare cases, which i could imagine, are mainly related to the actual data representation (i.e. linear light etc.), where other means (e.g. OpenEXR resp. floatingpoint values) make much more sense.

on my preferred systems, i.e. debian linux workstations, cineform support isn't provided at all by the precompiled ffmpeg-binaries/libs, but i never missed it in practice till now, because i really prefer to use more common solutions and their much better hardware acceleration support for all my practical work.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 7:27 pm

It has been reported few times as far as I know.
Same problem exists for ProRes. ffmpeg devs refuse to acknowledge that 12bit ProRes exists :)
Try yourself- do 16bit TIFF with eg. every 4 levels steps encode with AE/Resolve and decode by Resolve/AE and ffmpeg. You will see straight away that ffmpeg decoding looses precision.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 10:44 pm

it's a kind of wishful thinking, if one expects to preserve 'real' 12bit accuracy in a loosy codec, which uses some kind of YUV representation internally. as already mentioned, there are much more adequate other means resp. transport conventions available, if you really need this kind of accuracy in practice.
Offline
User avatar

Cary Knoop

  • Posts: 500
  • Joined: Sun Mar 12, 2017 6:35 pm
  • Location: Newark, CA USA

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostWed Jul 11, 2018 11:06 pm

One should not underestimate the need for bit depth.
Linear needs at least 15 bits to stay below the Barten ramp while log needs at least 13 bits.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostThu Jul 12, 2018 12:15 pm

Martin Schitter wrote:it's a kind of wishful thinking, if one expects to preserve 'real' 12bit accuracy in a loosy codec, which uses some kind of YUV representation internally. as already mentioned, there are much more adequate other means resp. transport conventions available, if you really need this kind of accuracy in practice.


Thee is no YUV "representation" as all modern intermediate codecs have RGB modes (and those are 12bit). Loss of precision due to compression is tiny- do you see much of a loss in case of uncompressed v. 10bit in ProRes etc? You will have this problem with relatively low bitrate h264/h265 encodes, but not for intermediate codecs.
Same applies to 10bit v. 12bit. Not sure why you so hard trying to "hide" simple limitation/bug in ffmpeg's decoding of ProRes and Cineform.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostThu Jul 12, 2018 7:26 pm

Andrew Kolakowski wrote:Not sure why you so hard trying to "hide" simple limitation/bug in ffmpeg's decoding of ProRes and Cineform.


no -- i don't try to hide anything about it!

this would be really absurd in case of free software, where everything is open to scrutiny for anybody, improvable by anyone capable, and usually accompanied by really harsh discussions and critics between developers.

i also explicitly mentioned my regret about this very annoying 8/10bit limitations of many open source video solutions.
but i'm also well aware about the practical reasons of this limitations and all the hindrances to overcome this legacy code flaws in the short run.

during the last months i was more involved with similar issues on the output side, where i also had to learn a lot about all this really unpleasant side effects of ABRG2101010/RGBA1010102 pixelformat compromises, constrained by legacy graphic code interfaces and hardware design. it's a horrible workaround, which will never work really satisfaying. the only consequent answer has to be seen in completely new designed system extensions, which handle this this kind of tasks resp. the more advanced demands of our days in radical different manner (e.g. https://www.x.org/wiki/Events/XDC2017/goins_hdr.pdf ).

but concerning your really fanatic insistence on this 12bit variants, i really felt remembered to those days long ago, when i was a young pupil at a technical school, and our teacher told us, how stupid it is, to just write down all the long row of numbers after the decimal point from our pocket calculator displays, instead of learning how to consciously limit the useful precision of our results in a much more adequate/suitable manner...
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostThu Jul 12, 2018 7:37 pm

It all has nothing to do with 12bit itself- its about ffmpeg not decoding ProRes or Cineform 12bit properly, simple as this (and fact that this has not changed in last few years). This is "bad side" of ffmpeg.
As you said- if camera/recording is good 10bit can satisfy most needs, but this is not the point.
Offline

Al Spaeth

  • Posts: 220
  • Joined: Thu Sep 17, 2015 9:48 pm
  • Location: South Africa

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 9:26 am

Hi Andrew - What about the mathematically lossless codecs like MagicYUV (which also claims to be the fastest codec)? Any better quality than the traditional intermediates mentioned above which we are more familiar with?
They claim "Unlike visually lossless codecs, MagicYUV is mathematically lossless, meaning that the decompressed output is bit-by-bit identical to the original input."

More info, color space, benchmarks plus an SDK here: https://www.magicyuv.com/

"This is what Cineform offers and it works perfectly in Premiere. No transcoding, no proxies- all is just a press of a button to jump from full res to half or quarter. I've tested 3x4.6K Cineform files in Premiere at 50% opacity - switch to 1/4th resolution and you can edit it on i5 dual core :)"

+1 Seems like fractional decoding can make 4K laptop editing a reality. Have you suggested it to BM as a future Resolve feature?
Resolve 14.1 free Win 10 64bit
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 10:41 am

Al Spaeth wrote:+1 Seems like fractional decoding can make 4K laptop editing a reality. Have you suggested it to BM as a future Resolve feature?


I'm sure they are aware of this possibility.
So far Premiere and Assimilate Scratch has it implemented. It's part of reference decoder, so not that much work is needed.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 10:56 am

Al Spaeth wrote:Hi Andrew - What about the mathematically lossless codecs like MagicYUV (which also claims to be the fastest codec)? Any better quality than the traditional intermediates mentioned above which we are more familiar with?
They claim "Unlike visually lossless codecs, MagicYUV is mathematically lossless, meaning that the decompressed output is bit-by-bit identical to the original input."

More info, color space, benchmarks plus an SDK here: https://www.magicyuv.com/



Lossless codecs are very different from intermediate. They work like zip/rar. They don't understand/need to understand video, as they treat stream as data and simply try to pack all repeated 0,1/patterns. They can be crazy fast, but also use very complex "packing" strategy (like ffv1) and then bit slower. Key issue is that you don't really know compression ratio until it's actually performed. Usually it's around 2:1, but can very from 1.3:1 to eg. 5:1 depending on source nature. All straight from camera recordings compress "badly" as they are full of sensor noise, so a lot of random data. Some anime compresses very well as quite often textures are simple (flat colors etc). BM lossless RAW compresses around 1.7x.
There is probably place for some lossless, but more video alike codec, which would do some frames comparison (so have temporal based features). I think AMV4 does this.
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 1:25 pm

many codecs have some feature that many people miss, feature that allow you to work fine and faster.
- scalability mean that codec could allow the software to read 1/2 1/3 1/4 and more data from original file without load too much memory and hard disk. you can find this feature in dnxhd/hr, prores, cineform and also in h264, but most of software (with except of original creator of codecs) use this feature.
this allow you to edit a 8k on a simple computer be cuase until you not need to see all 8k pixels you read only a 1/16 of pixels.

- fast encoding decoding algorithm to allow to edit write in easy hardware, in this feature prores and dnxhr are better than cineform be cause are faster to write and require less cpu effort.

- ability to manage more than 8bit datas... all pro codec like prores, cineform, dnxhr and also h264 allow you to write 12bit color depth files

the problem is not if we can use one or another codec, but also if NLE that we use can manage correctly the feature of codec.
think to active metadata of old commercial cineform license, today after 15 years we have same system only for red code workflow, on prores we have only some small info like camera metadata and lut no more.
we are here to ask codecs for cameras, but also a full support of codecs'features on Resolve 16 (i'm going to the future, resolve 15 is near next corner).
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 1:52 pm

carlomacchiavello wrote:many codecs have some feature that many people miss, feature that allow you to work fine and faster.
- scalability mean that codec could allow the software to read 1/2 1/3 1/4 and more data from original file without load too much memory and hard disk. you can find this feature in dnxhd/hr, prores, cineform and also in h264, but most of software (with except of original creator of codecs) use this feature.
this allow you to edit a 8k on a simple computer be cuase until you not need to see all 8k pixels you read only a 1/16 of pixels.


This is not so easy. For DCT based codes decoding at fractional resolutions is way more complex than for wavelet based (Cineform, jpeg2000). For intermediate DCT codecs this is doable (but only Apple has done in FCP X). Don't mix NLEs ability to preview at fractional resolutions with decoding source at fractional resolutions. This is not the same. Most NLE decode source at full resolution then drop it and do further processing at fractional, but this is not the same as reading source at smaller resolution. About all NLEs can do 1st one, but about none can do 2nd one.
Only FCP X for ProRes can read ProRes at eg 1/2 resolution. Other than this only Premiere and Scratch can read Cineform at fractional resolutions. I have not seen this in other NLEs/codec. Maybe latest Media Composer can do it for DNxHR, but I doubt. It's easy to check- just play files in timeline at different preview resolutions and check CPU usage. You need to see real CPU usage drop, not just eg 20%. For example in Premiere decoding 3K timeline takes 22% of CPU on my Mac, when preview is switched to 1/2 it's only 12%. Dropping further reduces CPU load also, but much less as decoding of Cineform becomes only tiny part of whole process of "playing" timeline. When you convert same file to half resolution and create Premiere timeline you get about the same CPU load as in case of 1/2 resolution preview of original file, so this shows you how cool is this feature. You have basically linear CPU load reduction (you need same disk speed though as files is always read from disk at "full resolution"). This is why you can work with 3x 4.6K Cineform files (with opacity) when you switch preview to 1/4 on dual core i5 in Premiere (it's in reality decoding 3x 1K stream).

For h264/5 decoding at fractional resolution is about impossible due to complexity of the codec ( maybe it's even impossible in practice at all).
Last edited by Andrew Kolakowski on Sun Jul 15, 2018 2:30 pm, edited 2 times in total.
Offline
User avatar

Micha Clazing

  • Posts: 126
  • Joined: Sat Apr 01, 2017 3:26 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 1:59 pm

Andrew Kolakowski wrote:For h264/5 decoding at fractional resolution is about impossible due to complexity of the codec ( maybe it's even impossible in practice at all).

What Premiere does with h.264 files is it disables inloop deblocking when scrubbing the timeline to significantly speed up decoding at the cost of accumulating error on inter frames. As soon as you let go of the timeline playhead it decodes the last frame properly with inloop deblocking and the artefacts disappear.
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 2:04 pm

Yes, this is part of h264 decoding spec- you can skip deblocking.
Helps scrubbing, but it still won't help with "full" decoding during playback/export.
Offline
User avatar

carlomacchiavello

  • Posts: 332
  • Joined: Tue Aug 28, 2012 6:04 pm
  • Location: italy

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 2:30 pm

Andrew Kolakowski wrote:This is not so easy. For DCT based codes decoding at fractional resolutions is way more complex than for wavelet based (Cineform, jpeg2000). For intermediate DCT codecs this is doable (but only Apple has done in FCP X). Don't mix NLEs ability to preview at fractional resolutions with decoding source at fractional resolutions. This is not the same. Most NLE decode source at full resolution then drop it and do further processing at fractional, but this is not the same as reading source at smaller resolution. About all NLEs can do 1st one, but about none can do 2nd one.
Only FCP X for ProRes can read ProRes at eg 1/2 resolution. Other than this only Premiere and Scratch can read Cineform at fractional resolutions. I have not seen this in other NLEs/codec. Maybe latest Media Composer can do it for DNxHR, but I doubt.


Hi Andrew, i wrote in the worst way but is the same that i want to write.
If codec allow nle to load a fraction of file is good, but if codec not have this feature you cant do it.
The good of cineform (which i use from 2002) is this feature that allow me to work on hd from ages, if not all nle or software use this feature is not cineform fault, is software's developers fault. :-D
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 2:34 pm

It's not fully developers fault either- it's limited by codec design and math behind it. Easy decoding at fractional resolutions is one of the key benefits of wavelet based codecs compared to DCT ones. For DCT based you need a bit of an effort to have partial res decoding and in case of complex codecs (like h264) it's in reality impossible/ crazy complex (so you don't have a single h264 decoder out there which can do it).

I would love to see it implemented for Cineform/ProRes in Resolve as in both case this is part of reference decoder, so it require just an integration, not a much bigger effort of writing actual fractional decoding process.
When I asked Assimilate they were already in the process of adding it.
Offline
User avatar

Cary Knoop

  • Posts: 500
  • Joined: Sun Mar 12, 2017 6:35 pm
  • Location: Newark, CA USA

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostSun Jul 15, 2018 4:11 pm

Andrew Kolakowski wrote:It's not fully developers fault either- it's limited by codec design and math behind it. Easy decoding at fractional resolutions is one of the key benefits of wavelet based codecs compared to DCT ones. For DCT based you need a bit of an effort to have partial res decoding and in case of complex codecs (like h264) it's in reality impossible/ crazy complex (so you don't have a single h264 decoder out there which can do it).

True.

One wonders why Wavelet based encoding is not more popular in DASH environments.
Offline

Al Spaeth

  • Posts: 220
  • Joined: Thu Sep 17, 2015 9:48 pm
  • Location: South Africa

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon Jul 16, 2018 9:26 am

Andrew Kolakowski wrote:
Al Spaeth wrote:Hi Andrew - What about the mathematically lossless codecs like MagicYUV (which also claims to be the fastest codec)? Any better quality than the traditional intermediates mentioned above which we are more familiar with?
They claim "Unlike visually lossless codecs, MagicYUV is mathematically lossless, meaning that the decompressed output is bit-by-bit identical to the original input."

More info, color space, benchmarks plus an SDK here: https://www.magicyuv.com/



Lossless codecs are very different from intermediate. They work like zip/rar. They don't understand/need to understand video, as they treat stream as data and simply try to pack all repeated 0,1/patterns. They can be crazy fast, but also use very complex "packing" strategy (like ffv1) and then bit slower. Key issue is that you don't really know compression ratio until it's actually performed. Usually it's around 2:1, but can very from 1.3:1 to eg. 5:1 depending on source nature. All straight from camera recordings compress "badly" as they are full of sensor noise, so a lot of random data. Some anime compresses very well as quite often textures are simple (flat colors etc). BM lossless RAW compresses around 1.7x.
There is probably place for some lossless, but more video alike codec, which would do some frames comparison (so have temporal based features). I think AMV4 does this.


Thanks Andrew - My codec knowledge is very basic but we ran some tests a few years ago using another NLE and MYUV. We decoded H.264 MP4 4K 8 bit files using MYUV (with help from the codec author). Decoding was very fast and filesize was about 10X AVC source. The NLE had no MYUV support but with the codec installed on a Win7 PC we substituted the VFW decoded files for the camera source files (same name) in the NLE. Results were amazing - smooth timeline 4K editing, scrubbing and playback on a cheap i5 with no discrete GPU at full 4K preview resolution! The same system struggled with 1/4 res MP4 proxies. I don't know of any NLEs today using 4K intermediates at full res that come close to that sort of performance. Realtime 4K AVC editing with 4K preview on an average home PC impressed me.
MYUV also has 10 bit RGB/RGBA and YUV 4:2:2 plus 12/14 bit RGB support which we never tested and I would assume RAW encoded files would have good timeline performance - but don't know .
I think MYUV is supported in ffmpeg and maybe Handbrake.

Surely there is some scope for fast lossless codecs for editing??
Resolve 14.1 free Win 10 64bit
Offline

Andrew Kolakowski

  • Posts: 3759
  • Joined: Tue Sep 11, 2012 10:20 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostMon Jul 16, 2018 9:55 am

Problem is that you don't gain that much. In some way 2:1 can be helpful, in other this is not much at all. With lossless you still need storage bandwidth about the same as with uncompressed. Decoding is fast, but that's about it. Intermediate codecs are much better compromise between quality v size v speed.
MagicYUV has 8bit+ support now. I worked with developer on it. We tried to make it a native codec for Adobe, but did not happen.
Offline

Martin Schitter

  • Posts: 730
  • Joined: Tue Apr 28, 2015 10:41 pm

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostTue Jul 17, 2018 12:36 am

Andrew Kolakowski wrote:For h264/5 decoding at fractional resolution is about impossible due to complexity of the codec ( maybe it's even impossible in practice at all).


no -- that's not true!
this feature is called scalable video coding (SVC) and is in fact available resp. specified for nearly any modern delivery codec. it's just not used and implemented in most practical cases, because it's only usefull for very specialized demands in practice.
Offline

Uli Plank

  • Posts: 2809
  • Joined: Fri Feb 08, 2013 2:48 am

Re: Cineform or DNxHD intermediates for editing in Resolve 1

PostTue Jul 17, 2018 1:02 am

So, the discussion is getting slightly academic here ;-)
Resolve Studio and Fusion Studio
iMac 2017 Radeon Pro 580 8 GB VRAM and 32 GB RAM

Return to DaVinci Resolve

Who is online

Users browsing this forum: Bing [Bot], Gary Hango, Morgisak and 25 guests