Show Posts

This section allows you to view all posts made by this member. Note that you can only see posts made in areas you currently have access to.

Messages - cmh

Pages: 1 [2] 3 4 ... 6
Well that's why I'm comparing apples to apples: MLVApp prores sRGB/rec709 compared to MLVApp cdng in Resolve with sRGB/rec709 in the Raw tab (but also other color spaces/gamma).
If you raise the exposure by +1.25 or 1.26 (or fix it in the tag with exiftool), you get pretty much the same picture (there's still -15 tint difference for some reasons, that's not the point tho).

I don't really want to talk about bmdfilm but it's more complicated than that (the raw tab is different between legit blackmagic files and other cdng).

Also I'll stop answering if a comment is ignoring the previous posts and just assume that I did something wrong. No offense.

edit: no adjustements made (other than + 1.25 exposure and wb in the raw tab), pretone curve disabled, default DaVinci YRGB.

I'm not entirely sure but to my understanding Movie Tab ratio is about liveview crop (it restrains the maximum resolution) but gives you the proper liveview and the raw video aspect ratio concerns the resolution, if not matched it might need a desqueeze in post. It is the way it is for technical and performance reason.
edit: matching both is perfectly okay.

And why is everyone so bothered by exposure? It's linear, so a compensation doesn't matter if you do it by hand, or if it's embedded in the file.

I'm not sure if you are asking how we came to talk about it or what are the technical reasons for those tests so I'll quickly glance over both so you don't have to search over the previous posts:

Context: Another ml user and I were sharing power grades involving cdng and bmdfilm and we both wondered why we had to raise the exposure by +2. He was afraid that it would increase noise. Doing some tests (checking highlight retentions with various MLVApp profiles) I realized that cdng exposure was always darker no matter the color space/gamut compared to encoded videos. Comparing ursa mini footage, the most obvious thing to pop up was an exif tag missing so I wanted to point this out.

Why it is relevant to me? If I mix Prores and cndg footage, I would only be able to correct cdng exposure. That's something to keep in mind before starting a grade.

That said, I don't care about ACR/Lightroom at all as it is not part of my workflow (I don't even understand why it matters but hey, if someone is asking for a test I'll gladly comply and it confirmed the missing tag). I found a discrepancy between prores and cdng. I don't care about a fix in MLVApp (exiftool does the job anyway) but I'm not the one qualified to say if it's a bug or a feature (that is up to you and your community to decide). Sometime those little bugs are the tree that hides the forest sometimes it's not worth fixing it for technical reason, I honestly have no idea. It's just an exif tag for cdng export.

edit: added MLVapp screenshot in a previous post as requested.
Also there's a lot of misconceptions around bmdfilm (in Resolve's raw tab, Gen1 is actually not a transform) but I won't go into there as this is not directly relevant to MLVApp (but if the option is present in MLVApp, people should be able to use it and expect a decent result).

Real quick (sorry for the squeezed ratio).


Rawtherapee (which is most definitely rendered as sRGB)


both really close to Resolve... so the BaselineExposure exif tag would be needed for Windows (maybe Linux) but not Mac OS for some reason. Very strange.

edit: added MLVapp

Yeah Togg from the forum and I had a discussion about a dpx (power grades if you are not familiar) and we both had to raise our exposure by +2 with it, so I'm pretty sure he's on Windows and would have the same issue. Maybe some other Windows users are willing to confirm the issue. What about Linux ?

Here's the dng I used if anyone needed it:

That said, I'll peace out for now.

This is definitly NOT what we got on Windows, weither it's in ACR/Lightroom or Resolve... I'm starting to think that it's more of a Mac OS thing now.

I'd say Resolve is closer to Lightroom in term of exposure (but with a wrong white balance, it has always rendered cdng like that and it needs like a -15 tint correction compared to prores) than MLVApp or prores.

Damn you are efficient and yeah I know, sRGB/rec709 and rec709/rec709 are really close (sRGB is just a bit higher) but I want to be as accurate as possible (I can see reinhard in your mlvapp screenshots tho and you extracted a frame you didn't export to cdng, just saying).


Prores from MLVApp in Resolve

cdng from MLVApp in Resolve

cdng from MLVApp in Lightroom

Give me a sec I'll do the same with sRGB/rec709 everywhere (since I think this is how ACR will display it, well I think).

I won't put that on laziness (maybe a lack of interest).
I don't know much about ACR but I do know that Resolve won't get the right exposure value for DNG produced without the proper exif tag.
Isn't ACR/Lightroom loading an embedded profile with lens data and color profile ? Then there's probably a way to strip that and compare to Resolve or UFRaw or RawTherapee.
Are those comparisons made with the same color space/gamut or just left with default reinhard on MLVApp for exemple (I did my video with rec709/rec709 in both MLVApp and Resolve for that reason)?
Are those previews an accurate representation of said DNG in the first place weither it's Resolve, Lightroom, MLVApp ?
At the end of the day if it revolve around more questions than answers, then I'll ask the real question? Is this even worth "fixing"?

Sorry for the late response.
I follow your tests. You converted a cr2 to dng and state it says 0.26 after conversion.
Right, I understand now.

Sure, it isn't an issue or wrong setting in Resolve?

I posted a quick video but there is no additional settings involved, it's not ACES, pretone curve isn't ticked (but I've tested, it doesn't change the exposure strictly speaking). The same exposure difference is for all dng files regardless of the profile applied given that you use the same cst/lut in resolve.

I don't have any other NLE installed but at one point I will probably upload a (boring long ass) pipeline comparison between MLVApp and Resolve on youtube, with various shot and go through these settings on screen so people can point out any mistakes.

If I find how to fix it I can put up a patch and fork the project real quick. Otherwise, my workflow will involve exiftool, no big deal it's extremly fast and more convenient than applying a receipt or use a dpx on Resolve.

Code: [Select]

I don't really understand your comment. Just to clarify the 1.26 exposure difference concerns all codecs not just proxies.
Anything above 0.26 will act as overexposure according to the original spec.
Adobe's specs? Do you know where I can find some info on that?

Sorry I always edit my posts without realizing someone replied, but both separated values,+0,25 and +1 could be export options, one added on top of the other.
Also, you're right, not only different nle has to be checked but there's brands other than canon, thinking about raw2mlv stuff.

It could be two explicit tickboxes at export or something, like "BaselineExposure 0.25" with an helper dialogue "match adobe dng converter exposure for canon cameras" and "add BaselineExposure + 1"  with the dialogue saying "match encoders exposure". Then I don't request anything just wanted to point out the discrepancies between various dng files in resolve.

The exact value has to be determined, 1.26 in resolve seems to match but I haven't checked other NLEs (baselight, vegas, adobe's products, apple's stuff, etc).

Edit:so far the tag matches various canon cameras but ofc it will be different depending of brands and models.

I did those tests, I got prores and h264 files (both rec709 profiles but I already checked other profiles, +1.26 exposure is still needed).
  • MLVApp cdng have a BaselineExposure 0 tag and need +1.26 exposure to match them.
  • Adobe dng Converter files have a BaselineExposure 0.25 tag and require +1 ish exposure to match them.
  • CR2 files doesn't have any BaselineExposure tag ofc and, imported directly into Resolve they need +2.26 exposure.
While not perfect (due to framing missmatch, color temp shift and probably debayering differences) 1.26 seems consistent for MLVApp generated cdng files.
edit: let me do a screenshot real quick. Well a video then it will be easier.
Code: [Select]

Most definitly but it's not gamma and color space related, I've tested alexa cineon linear bmd they all need that exposure fix. Let me check other codecs real quick (but I think we would have noticed it before if dnxhr or h264 lowered the exposure by more than a stop).

ACR converted files have a value of 0.25 (various iso tested).
edit: let me check what happened in Resolve...

Both the converted CR2 and the cdng needs a raise of exposure of 1.26 (approximately) to match untouched prores (rec709 tested for consistancy but whatever, it's the same for all of them).

Not yet but you're right the 1.26 value is eyeballed by comparing prores and dng files on the vectorscope.

It's bouncyball code, maybe I should pm him later to see if he's around.

I see some other stuff line 579:
Code: [Select]
        /* Baseline exposure stuff */
        int32_t basline_exposure[2] = {mlv_data->RAWI.raw_info.exposure_bias[0],mlv_data->RAWI.raw_info.exposure_bias[1]};
        if(basline_exposure[1] == 0)
            basline_exposure[0] = 0;
            basline_exposure[1] = 1;

I need to reinstall qt and a dev environnement, that will take some time with my subpar internet.

MLVApp do create tags (related to black levels, camera and lens info for exemple) in dng.c and I can see some reference to BaselineExposure in dng_tag_codes.h.
I'm not qualified to say wether it's feasible in MLVApp or not but I think there's good hope.

I did a quick test with kuchka94's settings (high framerate, 3x crop on/off, etc). Everything's fine.

I used to raise cdng exposure to 1.26 in the raw tab to match prores in Resolve but I realized that MLVApp leave the BaselineExposure exif tag at 0 compared to various Ursa mini shots I got where it's always set to 3.09 (and 0.76 on a Digital Bolex D16 according to the internet).
A quick fix that overwrites the original files with a value of 1.26, it's a pretty quick operation:
Code: [Select]
exiftool -BaselineExposure=1.26 -overwrite_original *.dng I tried few samples provided on the forum (7d 14 bits and 5d III 10 bits) and various gamut/color space: it's still 1.26, it seems consistent.

Thanks for the update.

Raw Video Postprocessing / Re: Powergrade for Resolve
« on: April 28, 2020, 05:59:00 PM »
I stumbled on some info on cdng and bmdfilm:

CaptainHook at Blackmagic Design
"Blackmagic Film" for Gen 1 is actually not a transform but passes out sensor space, or sensor RGB. You can basically think of it as "no colour science applied". So you would need a transform from that particular sensors response to the gamut of your choice which you'd need to get from that manufacturer. I believe Digital Bolex for example offered LUTs to transform from their sensor RGB to common gamuts like 709 via a LUT to be used in this workflow with DNGs in Resolve."
"Yes, DNGs in the metadata provide the matrices and AsShotNeutral tags needed to convert from XYZ to SensorRGB (and back) as supplied by the camera manufacturer so Resolve will be converting from SensorRGB -> XYZ -> Rec709. As for getting an XYZ output option I'm not part of the Resolve team so you would have better luck requesting that direct to them, but you could use CST to go from Rec709 primaries to any other gamut as the output from the RAW decode should not be clipped."

Jacob Fenn
"My issue ended up most likely being that the DNG's weren't relaying the ISO data needed for Resolve to properly debayer to Blackmagic's log curve. Therefore, when I'd apply what I thought was a normalizing transform via a CST node, the result was off because that CST node expects ISO 800. Thanks CaptainHook for help figuring that out."

To my understanding:
  • There is no information clipped if we use rec709 (or linear whatever) in the raw tab and it's totally okay to, before any grade, use the first node as a color space tranform to log (like rec709 to alexa or cineon).
  • Using bmdfilm in the raw tab is a no-no and needs further color processing (a lut to match this particular canon camera for exemple like Digital Bolex do) but to verify my statement we should both shoot a color chart (which I don't have) in the same lighting conditions and compare the results of our cameras with your 1.50 dpx for exemple.
  • There's probably missing metadata in our dng compared to blackmagic cameras cdng, like ISO apparently that would explain the fact that we have to change exposure (as we said before the whole histogram is on the left which is not what a blackmagic camera would get).

edit: It might be the value named BaselineExposure. Compared with exiftool it is set to 3.09 on an ursa mini, 0.76 on a Digital Bolex D16 and 0 on an eos m cdng from MLVApp.
Tf you want to overwrite the original files with a value of 1.25 so you don't have to correct for exposure:
exiftool -BaselineExposure=1.25 -overwrite_original *.dng
It is a quick operation.

Share Your Videos / Re: Canon EOS M RAW | Crane M2 | 22mm f/2
« on: April 27, 2020, 02:24:20 AM »
I love the videos but I also really like the cheapo lens hack, it's really cool. I mean, anamorphic for 60 bucks on amazon, common. I've seen hacked projector lens that are far less practical, ten times more expensive and barely sharper.
I'm pretty sure a 3d printed holder is achievable (I see some 3D print lens cap TinkerCAD tutorials on youtube). Getting that printed to a 3D print services in SF is probably super affordable.
Can you elaborate on the drawbacks, like focusing, how fast it is, etc ?

Pages: 1 [2] 3 4 ... 6