Cinelog - True logspace conversion for DNG and CinemaDNG footage

[Andy600]

@swinxx - The reason other devs produce so many luts is because they either 'creatively' produce them for each camera by color grading (not efficient and not accurate) or they include the colorspace transform in the LUT (not efficient but accurate - within limits).

We took a more scientific approach by giving the user access to transform to/from the colorspace as technical shaper luts in Resolve or OpenColorIO transforms in AE/PremierPro. This means we only need to produce one set of Looks and these looks are all created for a single colorspace - Cinelog-C.

Cinelog-C is a toolset for a raw (DNG based) pipeline and we haven't (yet) ventured into profiling baked-in looks such as Picture Styles, Cine-gamma, ProTune etc etc because all these are a moving target. It only takes the user to adjust a setting in-camera (contrast, color tone etc etc) to alter the output significantly and to maintain accuracy we would need to create input transforms to counter every possible combination of settings - this would be infinite when you think of all the cameras our there.   

The Cinelog-C OpenColorIO transforms can work in reverse on baked log footage from the Alexa, Amira, F5, F55, C300, C500 etc simply because we know the exact colorspace and log curve of each of those cameras. If you have even read a professional camera's manual you will note that the manufacturer explicitly states that the user should not alter contrast, saturation etc and should expose in a specific way when shooting anything with the camera's log profile. They don't want the user to affect the very technical color science that has been developed to capture and compress the full signal into a much smaller container - usually a ProRes file.

Cinelog-C is based entirely on open standards so that in can be used in a professional workflow. I know of 2 big budget movies currently in production where Cinelog-C is being used for DI because the DITs themselves are 100% confident in how it works and what it does - one of those films has several 5D mark III's (shooting ML raw video) in use alongside 2 ARRI Alexa's.

As for other LUT vendors and how they do things, well that's not for me to say. I know only 2 other companies that take every effort to ensure accuracy in their LUT/Look products and I know several that make claims that simply don't hold up under scrutiny.

[Andy600]

@Lars - correct but ACES's strength is also a draw back because it conforms every camera to a common look - ACES is great for camera matching (if the IDT's are accurate) and for VFX heavy pipelines but it can also dumb-down some of the aesthetics that make a camera like the Alexa unique.

ARRI are the only manufacturer so far to have followed ACES guidelines to the letter. It makes you wonder why Sony, Canon, RED etc don't.

[Andy600]

@Tullen

Up-scaling - Resolve
Debayer - ACR has the edge but Resolve is still excellent.
Noise reduction - Neat Video
Speed - Resolve
Color grading - Resolve

If time isn't an issue then ACR + After Effects will produce the best image but Resolve is amazingly versatile.

re: 50D - I can't take credit for anything ML related.

[Tullen]

Thanks Andy.

I guess Resolve sounds like a good option since it is also free, even though debayering of course is important. Would you say it is as good editing software as Premier Pro? As I have understood ACR is very powerful to recover details in the images, would it make sense to use ACR for debayer and highligt rec etc. and then continue in resolve? Would it be best to stay with the raw files the whole way or would you recommend DNxHD? I also don't see you mentioning Cinelog in this workflow. Maybe you think that was obvious, but would be great to hear how to integrate it in the work flow proposed.

Well, you where very active in the 50D thread, supportive and always fast with a knowledge answer, so even if you did not work on the build except building it your contribution was very much appreciated. 

[Andy600]

@Tullen - I'm not an editor but Resolve's editing capability is pretty good if you don't have Premier, FCPX, Avid etc

A start to finish raw workflow is technically the most versatile but it's also very processor intensive (i.e. realtime debayering) and really requires a good RAID setup to handle the data flow. In reality it makes much more sense (economically and timewise) to develop high quality 10 or 12bit log masters in ProRes or DNxHD from the raw footage because these are much smaller file sizes in comparison to the raw source material and there is no debayering required for playback - You can edit/playback in realtime and use more plugins, luts etc even on a modest machine - that's where Cinelog-C comes in. We detail some of the benefits on the website.

The other benefit is Cinelog-C colorspace. We provide a common set of REC709 and graded, filmic look luts (with more to come) that are specifically developed to work with Cinelog-C footage. A Cinelog-C ProRes intermediate is basically the same as a Log-C ProRes intermediate that you would get from the Alexa and Amira cameras but Cinelog-C has a slightly different log curve that's more suitable for the dynamic range of Canon DSLR's, BMD cameras and a few others - plus it's the same transfer function (Cineon) used in film-to-digital and digital-to-film pipelines - and because you are generating the ProRes or DNxHD file in post from the raw data (as opposed to baking it in-camera) you can select the quality, bit depth and codec of your choice plus correct white balance and exposure if needed. A Cinelog-C encoded 4:4:4 DNxHD or ProRes file can handle some serious/extreme color grading and will look the same as if it was graded raw.

[Tullen]

@ Andy

As I said, I can go either way with what program to use, whatever gives me the best result. Obviously the biggest issue right now will be my skill, but if I am to learn I rather learn the best way straight away. I do however understand that your advice is regarding the picture quality and not how good the programs are from a editing point of view. I have a i7 2700k at 4.5 GHz (effectively 8 cores), only 8 gig of ram, Geforce 670GTX and SSD hardrives for editing. Not in raid though, maybe something to plan for the future.

I will film a lot of fire art performance (at night with the fire as main light source), if that matters. But of course, if I get as good result with a faster and lighter Cinelog-C DNxHD work flow, I see no point in staying in Raw the whole way.

When I read on your website the process looks a bit different for Cinelog and Resolve compared to Cinelog and ACR. My impression is that the ACR one is more advanced and will give you better result. Is this correct or am I interpreting it wrong?

[Andy600]

@Tullen - re: highlight recovery in ACR and Resolve

As I understand it, ACR attempts to rebuild some spectral highlight information (ultra brights) that may be missing in 1 or 2 channels (RGB) by copying/averaging/borrowing information from a channel that does have information.

Resolve Highlight recovery simply adds a shoulder to the uppermost part of the log curve and adds linear gain above the shoulder to increase perceived brightness and make it look more lifelike. It's basically a contrast function limited to the upper part of the signal.

[Andy600]

@Tulen - Fire Art at night - I take it you like a challenge then 

If there is little or no ambient light you're looking at 15+ f-stops of scene dynamic range so you'll need to decided about what to clip when shooting with the 50D's ~10 F-stops DR. I suggest shooting at a minimum of 800ISO (the max you can really go to on the 50D) and expose for the fire artist - use some lighting or bounce if possible or your shadows will be mush.

The ACR and Resolve versions of Cinelog-C use different techniques to achieve the same end result. Technically speaking, the OCIO transforms that come with the ACR version are more accurate because they don't use a 3D lut component for the linear colorspace transform (they use RGB matrices instead) but they are visually no different. 3D luts always introduce tiny mathematical rounding errors but you wont see them in the image - It's the same for any 3D Lut format regardless of the cube size and not something to worry about.

[Andy600]

For anyone that's interested in learning more about colorspaces, luts, log vs linear, OpenColorIO etc I suggest reading this: http://github.com/jeremyselan/cinematiccolor/raw/master/ves/Cinematic_Color_VES.pdf

 

[Kharak]

@Tullen

If you want to retain maximum quality and still get real time playback when editing.

I found that using proxies is the way to go.

step by step:

1. MLV to Cdng (Raw2cdng)
2. Smart import script for AE (do a search on the forum) Will make a composition for each image sequence and add it to render queue (set default render format before you do this)
3. White balance and apply Cinelog or Visionlog in ACR (Do this to avoid having to go back again after editing to apply the Cine or Visionlog)
4. Render all clips to h264 - 5 mbits or lower if you feel so. (Remember to keep same file names/reel names of your proxies and image sequences)
5. add proxies to Premiere and make a Track named 'proxies' just to keep things separated
6. Add the AE compositions to a track above the proxies (preferrably name the track RAW) and Group each Raw and proxy clip together
7. Make the RAW track invisible so that you only cut the Proxies and if you've correctly grouped the RAW and Proxies's without first making any cut to either one, then the cuts you do to the proxies will be doing the same to the RAW footage. You will basically cut and playback RAW in realtime on a very modest computer.
8. Go back to AE and colour grade add vfx or whatever you feel like, because the AE composition is grouped with the proxies, the lenghts will match and all you have to do is disable the Proxy Track and enable the RAW track and voila! You are ready to render a full quality image out of PP

Important: Make sure the compositions are exactly the lenght of each DNG sequence before rendering to proxy or else your footage wont add up when you group them and you will have black spots on the timeline.

Also because you've added Cinelog or Visionlog to your footage before rendering, the proxies are flat and very unimaginative to edit. Just to help you better get a feel when editing, I would apply auto contrast or just some easy "look" with high contrast to better give a feel for what you are doing. Don't start colour correcting the proxies, it serves no purpose at all




Note: I wrote this completely out of memory, I haven't had a computer for more than a month now as mine suffered some horrible electric meltdown So maybe I missed a step on how I usually go by my workflow, but it has worked wonderful for me, retaining all information because of no loss in conversion, from start to finish.

Sorry for taking this a bit too much off topic.

[Andy600]

@Kharak - I agree to some of what you say but proxy workflows are really only useful these days if your working on a low powered laptop or in UHD/4k on a low spec computer.

A ProRes or DNxHD log intermediate removes the need for proxies. It is efficient, editable and still allows scope for color grading. I know some still use proxies but it overly complicates a workflow more than is needed. If you have properly white balanced your footage and corrected any exposure offset there is no point in staying in the raw domain because all you are doing is deferring debayer.

A ProRes XQ or DNxHD 1:1 file has all the wiggle room you need - even for extreme grading and VFX. The difference between the raw and ProRes graded output is tiny and, unless you are going to DCDM you will likely end up delivering in 8-bit H.264 anyway.

[Kharak]

yes Prores and DNxHD are more than enough in almost all cases.

But for personal projects I like being able to jump back in to ACR and tweak highlights and shadows if I see so fit on a later bases (which I pretty much always end up doing). But if I am handing footage over to someone else to work on, I give them DNxHD or prores files. CDNG is in most cases too big a load, especially if they are not used to working with it.

But for my previous tutorial, for DNxHD/prores follow the steps to step 4. and instead of H264, set it to DNxHD/Prores and you are good to go without the hassle of proxy

[Andy600]

@Kharak - Please tell me you're not using the highlight and shadow recovery controls in ACR on any important/commercial projects

I had a client last year who did a short broadcast piece for the BBC (filmed in ML Raw) and it failed the Harding FPA test which tests for flicker, flashing, patterning that can cause epilepsy - It was very subtle but we eventually discovered it was the dynamic highlight and shadow recovery in ACR that was causing the flickering and that's why we say do not use those controls for video work.

[Kharak]

@andy

Yes I do fiddle with highlight and Shadow slides in ACR even after applying Visionlog, I mentioned this to you few pages back when I was asking about how much shadow recovery Cinelog had vs visionlog and btw I could see your Cinelog had more shadow recovery but more saturation (no biggey concerning the saturation from my part). I have yet to buy Cinelog C because as stated above, my computer had a mental breakdown and I've been without a workstation for more than a month now. But I will be sure to invest in Cinelog C as soon as I get my hands on a new computer, most likely next week sometime.

First of, increasing shadow recovery and highlight recovery in ACR does not work for every scene, I am very aware of the flickering that ACR can induce as it processes each individual frame as an independent photo.. But I have also learned that ACR focuses its adjustments based on the brightest highlight in a scene. So that in many scenes you can actually increase the shadow slider without introducing flicker. You introduce noise by doing this, but its nothing I haven't been able to fix/hide with noise reduction and colour grading and thereby increasing the dynamic range beyond what VisionLog was able to offer. Yes in cases I have been forced to keep all sliders on 0 (except exposure) because it would introduce flicker because of highlights changing or being blocked.

Scenario:

You are shooting a talent, you set up the scene and lets say there is a small specular highlight or just some lamp in the background (our brightest spot in the scene or bright enough to possibly cause flicker). As long as the talent or whatever else might be going on in the scene does not block or "provoke" the Highlight throughout the shot, you can pull and push highlight/shadows as much as you want without it causing any flickering, but sometimes I still get minimal flicker, but the flickering will be so minimal that it will fall natural to the scene and without the audience noticing. When I mean minimal, I mean not actual flicker, but supple change

Its trial and error in most cases.. But 90% of the time I try and pull shadows up in ACR and see if I can get away with it and for the most part, I do. As said before, I always push shadows cause I feel that VisionLog does not do a good enough in that part.

You explained the science behind the highlight recovery of ACR vs Resolve a few posts back, I didn't have the explanation in mind, but have all the time stated the same that ACR's highlight recovery is superior to Davinci.

[reddeercity]

@Kharak , +1 on your post. I agree with you on the usage of ACR in A.E. As I also use hi-light & shadow recovery.
Without any problems with flicking , I do not use Cinelog C or any Luts or transforms to log, just flat out the image
and tone curve to legal Video Colors. But I export to the new ProRes 4444XQ which is reported to be linear , (I'm still trying to confirm this) .
My two Cents 

[DeafEyeJedi]

+1 @Kharak's post regarding AE... To be able to use ACR is a no brainer. Gotta love their HL & Shadows recovery.

Even works well with Cinelog -- still trying to get the hang of the workflow though In order to get the look I want!

Thanks again @Andy600 for keeping us all well informed and up to date.

[mothaibaphoto]

+100 reddeercity.
ACR aware of not to flicker as long, as it run in AE, not LR or Photoshop.
I'm on windows, so i export to tiff sequence and encode to Prores with ffmpeg.

[Andy600]

Guys - If you're not seeing flicker then great! - but it can and does happen and is a pain to remove later. It usually occurs when a bright object enters the scene. It can be very subtle but I have seen more extreme flickering that looks a little like auto-exposure or AGC. It can be less noticeable on a computer monitor but much more visible on a modern Plasma or OLED display with a higher contrast ratio and, if you do any work for a reputable broadcaster it cause your work to fail the Harding test and will not get broadcast. I know most users here do not produce for broadcast but it's something to bear in mind if you ever do.

Cinelog is for a professional workflow and you should follow the same guidelines that apply to all in-camera log profiles i.e. Log-C, S-Log, Canon Log, REDLogFilm in that you should not mess with the curve at this early stage (The only exception is if you are going from raw to a deliverable output - i.e. not creating a log master).

Moving the Highlight and Shadow sliders in ACR will introduce an S-Curve to the log curve and you will move the measurable black point, white point and 18% gray points. ACR does not have the tools to tell you where you have moved them to.

The V3.0 profiles in ACR do a very specific job - they make it possible for ACR to pass up to 13.5 F-stops of scene linear (HDR) information to the host (After Effects, Premier, Photoshop) using a well defined log curve that can later be remapped to exact scene linear values - VFX artists and colorists will understand the importance of this.

Ideally you could debayer and transfer scene linear code values (not to be confused with Linear gamma) but ACR has a bottleneck. It does not use input-to-output floating point processing. It's limited to 8 or 16bit output and HDR values extend beyond what ACR can transfer without curve manipulation (the adjustments are not internally measurable) and so data outside of ACR's output range gets truncated - this is another reason for the log curve. Another thing to remember is that ACR to After Effects only transfers sRGB primaries.

When I use HDR in the above description I am not talking about HDR in the sense of a look. HDR in this case simply means dynamic range that extends beyond the range of what can be transferred or displayed.

If you don't alter the log curve in ACR you effectively have a Cineon log representation of scene linear vales in ACR. Add the Cineon converter plugin to this with the settings: log to linear, white level 685, black level 95 and gamma 2.39 and you have a full range REC709 signal (well, rec709 has a tiny linear portion at the foot of the signal but it's close enough). Change the gamma to 1.0 and you have linear gamma etc etc. Add Cinelog-C to Scene Linear LUT and you will see all your superbright values extending beyond normalized 0.0 - 1.0 output (but you must use 32bit processing in AE to see this). Also, the exposure plugin in AE will behave!

Part of what I'm trying to do here is to make it possible for a professional DI workflow (for producing log masters from your raw images) in a tool that was not specifically designed for that purpose - and, at the same time, help you understand a little of the terminology and practices of professional DITs and colorists so that if you ever work with them you will know a bit of their language. It's not just about marketing our products. 

If you don't use Cinelog that's fine too. I hope the information is useful 

[Lars Steenhoff]

Andy,

I'm using a workflow to adjust my files in photoshop and the results are great.
I'll tell you the steps and please let me know if I'm doing it right ?

- I shoot the canon 5dmk3 and push the rate button to set a white balance measure. ( while pointing at a gray card )
- I set the exposure using the ML spot meter to check and expose as much as possible to the right.

1. First I convert the MLV to DNG with MLVFS. ( with bad pixel correction on and stripe correction on )
2. In Photoshop I import the DNG in ACR
3. I change here the lens corrections ( remove chromatic aberrations )
4. Use cinelog v3.
5.  Sharpening in ACR at 0
6. I use native resolution 1920x1080 or I use 200% in ACR to get 3840 x2160
7. I use 16 bit file
8. I use srgb ( I'm wondering if I should use pro photo rgb or color match rgb? )
9. adjust white balance if needed
10. adjust exposure of needed

11. The file is now opened in photoshop as smart layer so I can always go back to adjust if needed.
12. Apply adjustment layer : Color lookup
13. Select Cinelog v3 to cinelog-C.cube
14. Make another Color lookup  adjustment layer on top of the previous one
15. In this Color lookup layer I choose the Look I like, for example:
16. [CLF] Kodak Vision Premier 2393 grade 05.
17. Use sharpening if needed, this can be in the ACR, or on the DNG smart layer with a sharpening plugin.
18. Use noise reduction if needed, again on the smart layer under the Color lookup layers.

19 Export to pro res or still from photoshop.
20. This seems like many steps for a still image but this can all be automated when setup properly.
21 I know the same could be done in after effects and would probably look the same or better, but I just like to experiment with looks in photoshop.
22. I'm using this workflow also for full resolution stils from the 5D, ( 22 megapixel ) and it gives me a lot of control over the colors I get from my still pictures,


One more question do you also have .cube lut for Cinelog to BMCC film?  I have a couple of looks for BMCC and would like to use them on the cinelog files.

[Andy600]

Hi Lars,

Yes, that workflow will work fine. The ACR version of Cinelog-C is built to work in a REC709 workspace (i.e. it's why we say to choose the REC709 workspace in AE project settings). REC709 shares the primaries of sRGB so setting ACR to sRGB will produce the correct colors. The difference between sRGB and REC709 is the transfer curve and gamma. sRGB is 2.2 gamma and REC709 can be anywhere between 1.9 and 2.4 depending on the display device. It won't make much difference to work in sRGB. ProPhoto has a different white point and lower gamma than REC709 and sRGB so it's not very useful for video work - there is a transform to/from ProPhoto colorspace in the OCIO config though

re: a Cinelog-C to BMD Film lut for Photoshop, AE and Premier Pro. The transfer functions are incompatible because BMD Film has less DR than Cinelog-C and Cineon. Technically it will clip the output but should still have enough DR for EOS cameras.

You can actually generate your own transform LUTs using the OpenColorIO plugin in AE or PP. If you want to go straight to BMD Film without transforming to/from Cinelog-C just select Cinelog V3.0 as the input space and BMD Film as the output space then click export. I think a 3DL format lut will work in Photoshop but I haven't tried it yet.

For BMD Film it would be better to make the Photoshop workspace BMD Film colorspace using an ICC profile. We need to generate this and it's on the list of things we plan to do.

[Pyriphlegethon]

Andy, will Cinelog have a presence at NAB?

[Andy600]

@Pyriphlegethon - Sadly not this year. NAB is a huge expense for very small companies like Cinelog, especially as we are based in Europe. If this year goes according to plan we should be at NAB 2016. I'm not sure if we'll be exhibiting but I should be there.

[Lars Steenhoff]

IBC ?

[timbytheriver]

@Andy600 BSC Expo Pinewood?

[Andy600]

We're too late for BSC this year but IBC is a possibility. I certainly need to get out more