massive shift in green color on BMPCC4K

So is Resolve Color Management. So are normalizing LUTs as specified by camera. Who knew this was such a mystery.

So is Resolve Color Management. So your normalizing LUTs designed by camera.

Whatever. Tried them all and tested them extensively but ACES works best for me ( but what do I know )

This is my last comment here. Whoever wants to listen, they can listen.

Paines: What I'm saying is the method is being promoted without evidence of efficacy, by people who haven't demonstrated basic competence in the barest essentials. And/or by persons claiming defects in traditional grading which reflect ignorance, not the methods themselves.

You are reaching, big time. And you're been condescending, and very, very rude.

Griffith: Because it's specifically designed to fix the issue we have here which is getting everyone so worked ie matching multiple camera types on the same timeline.

Exactly. While I didn't use the ACES method, ACES has a massive color space too, and it should provide us with more accurate colors when we compress them back to rec709. Hopefully.

So, I shot with the BMPCC 4k this weekend, so I have some fresh footage. The result of grading using the 'traditional' method, is UNDERWHELMING. And it's not lack of knowledge. I know how to grade, I've been shooting official music videos and editing them and grading them for more than 10 years now. Again: there is usually a yellow color cast when the green pixels of the sensor get oversaturated, because Resolve doesn't know how to fit these colors properly in rec709, without clipping them and shifting them. Their "compression" method from BRAW's wider gamut to rec709 is lacking, DESPITE their 32bit float engine.

For some reason, the same problems do not exist if you color grade as if with an ARRI Alexa. This was true when Juan Melara (who is a professional colorist btw) did his video in 2017, and it's true today. It's possible that Arri's LOG footage is either wider, OR, there is special code in Resolve for Arri specifically (maybe they used some of Arri's recommendations or SDK code). Suddenly, by linearizing the footage and turning it to Arri LogC, and grading within the Arri log space, the colors are vibrant naturally (no saturation slider on my part), and there is NO cast (essentially, there's no color shifting of green to yellow, or blue to cyan etc). There is perfect color separation when you grade that way.

I've tried grading with and without the Melara method, and there is a massive difference. The Melara method holds more saturation on the same amount of contrast. As you would expect from a film curve, in fact. Only changes I made was to use Rec2020 instead of rec709 in the RAW tab (and the equivalent changes in the nodes afterwards). Then grade in between as you would normally (I used a Kodak Ektar VisionColor emulation lut, FilmConvert, 2 more nodes of color adjustments before and after the lut/plugin, and at the end, 4 nodes of texture adjustments, e.g. grain etc). At the end, what comes out is rec709, but it's a rec709 that's not as lossy (in terms of color accuracy as far as the human eye is concerned) than you would get if you grade traditionally.

The results talk for themselves. Watch in a 4k TV via Roku's Vimeo app if you can (or a good PC monitor). The color is perfectly separated, neutral, and naturally saturated. Perfect skintones too, Arri-worthy!

I'm not ever going back to the default route of color grading. As far as I'm concerned, I found the solution to the problem that was bugging me since I bought this camera. The solution was emulating Arri.

The results talk for themselves. Watch in a 4k TV via Roku's Vimeo app if you can (or a good PC monitor). The color is perfectly separated, neutral, and naturally saturated. Perfect skintones too, Arri-worthy!

I think you have posted the wrong video as I can't relate the above description to what I'm seeing.

I found the solution to the problem that was bugging me since I bought this camera. The solution was emulating Arri.

i founded me too that work on bmd shooting (raw and prores) like arri log all work better, simply to my taste is better rolloff of highlights, color are gently and smooth, no oversaturation of many color.
Before to hit with pocket4k, i can tell about bmpc4k and ump g1, if i mimic that is arri color all taste better (to my taste, obviously).

Rec.2020, ARRI log, RED wide gamut (which I used successfully at Dmitry’s suggestion when I had problems controlling saturated red LEDs): different paths but similar rationale, no?

DVR, RCM, ACES: choose your weapon when you understand how each work.

Yeah, I was a bit rude, but there is -- grading preferences aside -- a factual claim here we could look at: Eugenia writes that "there is usually a yellow color cast when the green pixels of the sensor get oversaturated, because Resolve doesn't know how to fit these colors properly in rec709, without clipping them and shifting them."

Here's lots of green, thanks to Resolve Color Management and rec. 709. I left the high saturation, typical of RCM (much less so, with ACEs). There's no color correction here per se:

green_1.18.1.jpg (700.27 KiB) Viewed 14309 times

Any evidence of "clipping" and "shifting" of greens? Anything on the order of the jsfilmz yellow/browns in lieu of green?

Rec.2020, ARRI log, RED wide gamut (which I used successfully at Dmitry’s suggestion when I had problems controlling saturated red LEDs): different paths but similar rationale, no?

Yes. The wider gamut helps because it seems that Resolve uses Arri and RED's own software libraries when using their footage, as provided by these manufacturers. And these manufacturers have done their work to make sure they map the RIGHT colors from their wide gamut color space back to rec709.

DaVinci Resolve's DEFAULT engine does NOT do that extra-step work (despite their 32bit float engine -- it's a matter of algorithm). So when using any other kind of footage, not only BRAW, but even S-log or v-log, Resolve is going to use its default "compression" engine (that "compresses" the colors from wide to rec709) which is VERY lackluster in terms of being smart about it. The result is color shifts, casts, and clipped colors, with an "anemic" low contrast look.

That's why turning the footage to ARRI or RED works. Because when you do that, internally Resolve uses THEIR libraries, not Resolve's defaults. And it works even better if you first linearize your footage to turn it to Arri or RED (so having RAW footage helps heaps on that regard).

I made extra tests the other day, turning s-log, v-log and f-log to Arri too. Every single of these cameras BENEFITED from turning into Arri's space, because I was fooling Resolve to use Arri's more accurate algorithms for rec709 mapping. Canon had the least problem, since I didn't use log (it was true rec709 footage). But the moment you're using log or raw, and it's not ARRI or RED, so you're forced to use the default Resolve internal method, you're in trouble.

Blackmagic should fix this problem in Resolve. Same goes for Premiere btw. It's not smart about it either. Both their remapping algorithms suck.

The 2019 X-Rite Color Checker has a green patch that is pretty close to Ryobi green, at least to my eye.

I put a link to the Ryobi Power Washer I have with a color chart in a new BRAW clip. It is Q0, the rest of the details are in the clip? The clip was taken on a loading dock under a pavilion to take away any harsh shadows. I did a custom white balance off the color checker in camera.

It's a default YRGB project in Resolve, I then added the Pocket 4K V4 Lut, then in the color wheels tab I increased saturation to 80, set Hue to 48 and Gain to .81 This where i think it is good match for Ryobi's product images as shown on their website.

Ryobi Green - Hue Adjusted in standard YRGB and Pocket 4K Lut

ryobi-braw-hue-adjusted_1.49.1.jpg (471.94 KiB) Viewed 14305 times

Here is the link to the original BRAW clip in Q0.

https://we.tl/t-fsz5SwYtG1

Any evidence of "clipping" and "shifting" of greens? Anything on the order of the jsfilmz yellow/browns in lieu of green?

Yes! It has massive issues of yellow color shift. The whole image looks yellow (as if you put a slight yellow filter in front of the lens). The saturation and contrast are unnaturally bumped too btw in order to get it to kinda behave in rec709. You can try yourself Melara's method (it will only take you a few minutes), just make sure that where he says "rec709" in the raw tab and 1st node, you use rec2020. The rest is the same. You will see a difference, achieving true color separation and not having that yellow cast.

I think it's called sunlight ... and some highlights I let clip. Here it is, with the WB dropped about 1000 degrees:

wb_2.2.1.jpg (555.33 KiB) Viewed 14316 times

No, because the problem exists in the shaded areas too (it's just more pronounced in the sun areas). There is a cast in the whole image. Please try his method.

Edit: I saw your edit with the WB fix. It looks much much better. But 1000 points of WB fix is a lot. it would have made everything blue. Sure you didn't use Melara's method?

Nope, no Melara. In the original version, the initial WB setting, which was the 5600 daylight default, combined with the very high saturation gives it that intense cast. My intent was to show that the greens "held", unlike the jszfilms stuff.

Anyway, the initial WB setting was probably off, though not by as much as 1000 degrees, since the correction went a bit teal. But when 5600 was added to the super-saturation, the highlights you object to got accentuated.

Anyway, both versions, and versions I didn't try to cook up, are all available with the usual tools. I did fool a bit with Melara on another shoot and in all honesty, didn't see the advantage to it. But I'll try to give more time.

Thank for that further explanation, regarding potential use of ARRI libraries, Eugenia.

I'm not going to argue this one any longer, but I don't think the claims which Eugenia is offering for the interpretation of BMD log have been established by the proofs she's cited, or the samples of her own method that she's provided. There are many blanket statements here ("anemic", etc.), unsupported by actual example. Much of the basis for this line of argument was already retracted by Juan Melara himself.

No one disputes that there are differences between LogC and BMD Film normalization-- highlight roll-off, for example, which is no secret -- but I have yet to see any proof that these differences are intractable in post, or that there's anything to be gained by forcing BMD footage into LogC space.

All this is complicated by wholly different expectations (and tastes?) in color grading. I wish someone from BMD would comment.

Question though regarding the references to a linear space. What is that? Where and how would I take BRAW and put it in a linear space? Thanks.

I've tried also to put my clip through color transforms to Arri and RED and I still have to adjust the hue to make it color accurate to the referenced product image.

The clips presented that were transformed with the alt method seem too cyan to me, really blue / cyan.

For me, I see that the Pocket 4K has a slightly less accurate hue than the Canon M50 or DSLR. I happen to own the Pentax 645Z so I can't directly test it against the Canon. The Pentax 645z BTW is also slightly less accurate than an M50, so what?

Pentax 645Z https://www.imaging-resource.com/PRODS/ ... 645zA5.HTM

Canon M50 https://www.imaging-resource.com/PRODS/ ... -m50A5.HTM

I'm skeptical of changing the color space to Arri or RED to achieve a more accurate hue in the Pocket 4K. The hue adjustment tool built into the color wheels tab is already doing it for me.

Eugenia, I don't understand why you keep saying the greens are clipped as if it's a problem with anything other than your work flow. They aren't clipped because of the camera or anything like that.

YOU are either clipping them or causing perceived clipping by going into a smaller color gamut from raw tab. The bmd color gamut is larger than rec709 or rec2020 so you are causing the problem yourself in the grade if you see color clipping in greens It has nothing to do with the sensor, it's your methods that aren't working well with the footage. The BMD gamut is huge and is best worked in for full color values either in it's native color space or something as large such as RWG.

If you want to work in a smaller color gamut you need to use the proper tools to map them in your node tree if it's not working correctly with your current methods.

The gamma from bmd film can be mapped however you want and can look identical to Arri because if the data is there to do so. The sensor is perfectly suited for it. Pushing to log c might be advantageous if you like working in it but isn't required for good roll off. What is required is proper correction/grading.

Also on the shaded shot... I'm not sure about you but usually when sunlight peeks through like that I expect a little yellowing since shaded areas tend to be bluer than areas hit directly by sunlight. The warmth looks very pleasing, however if you were trying to achieve a greener look, as shown by John, it's perfectly achievable if you grade it.

It is less about color spaces but more about sensor factory calibration. As it was described many times before - BM cameras just capture less greens in some situations. This problem consists of sensor hardware, sensor glass, input color space for RAW, software processing tools. Companies like Arri, Canon or Fuji have 100 years background history and so way better understand things like film color, glass filters chemistry, light spectrum and can provide better calibration. BMD is more like young company with tech background, who developed its first camera 7 years ago and priced it $3000 instead of Arri $60000. I guess BM do that they can do within their resources and knowledge, and they and always learn on mistakes.



As result Pocket 6K capture way more vibrant greens compare to all older cameras, hope other BM cameras will do the same in future.

Color Science gen4 produce "happier" look without additional color correction, but similar to BRAW it is more like compromise option that is far from real RAW. Gen4 is very limited compare to Gen1 in BMCC,BMPCC,BMMCC cameras and don't allow to use native sensor color space. Same goes to Ursa4.6K Color Science Gen3.
BMCC,BMPCC,BMMCC with Color Science Gen1 use untouched native sensor color space and so provide more freedom for color manipulation and correction outside RAW.
I shared some tests here viewtopic.php?f=2&t=114403

Good comparison of the BMPCC4K and BMPCC6K in terms of colour. The sharpness comparison though seems to be more of a difference than I’d expect. Too late now, but that would have been a good place to show the BMPCC4K without the SpeedBooster XL since I’m suspicious that the XL might be softening the image more. I wonder if anyone has the XL and the Ultra and has any conclusion about the sharpness of both versus no SpeedBooster?

Good comparison of the BMPCC4K and BMPCC6K in terms of colour. The sharpness comparison though seems to be more of a difference than I’d expect. Too late now, but that would have been a good place to show the BMPCC4K without the SpeedBooster XL since I’m suspicious that the XL might be softening the image more. I wonder if anyone has the XL and the Ultra and has any conclusion about the sharpness of both versus no SpeedBooster?

I had a viltrox speedbooster and a Metabones at the same time. And I actually found the metabones to be softer.
I'll check on my 4k to compare viltrox vs no viltrox with one of my takumar lenses. Just need to see if I have a 42 to mft adapter.

Same time even Pocket 6K capture is less greens than RED camera. I recommend download original RAW filesfrom Youtube post and compare in Resolve manually.

Oh, man, Dmitry. This same youtubber -- again????

The BMPCC 4k/S1H comparison, with his unaccountable BMPCC 4K results, and the previous Jack dispute, who also relied on him, didn't discourage you?

I think maybe we need more reliable testing sources, before drawing conclusions about much of anything....

I'm honestly not trying to be difficult about this, but as seen in this thread and many other situations, regardless of the weaker greens on capture compared to other devices, the same looks and color science of other cameras can be achieved rather easily with correction. If you don't like the out of the box look, all that is required is correction. Whether it's gen1 or gen4, there is enough data in the color space to correct to the standard rec709 in any look you wish to achieve.

That correction can be developed into a lut or powergrade or LMT for instant results for other projects. I am having a very easy time matching looks across cameras with ACES and LMT's.

I get that out of the box things don't look the same as other cameras, but its correctable.

I'm honestly not trying to be difficult about this, but as seen in this thread and many other situations, regardless of the weaker greens on capture compared to other devices, the same looks and color science of other cameras can be achieved rather easily with correction. If you don't like the out of the box look, all that is required is correction. Whether it's gen1 or gen4, there is enough data in the color space to correct to the standard rec709 in any look you wish to achieve.

It's true, you can adapt or correct the hue if needed in default settings.

I agree with Eugenia that there is hue shift when saturating the Pocket 4K color in the default REC 709 timeline and YRGB color seen below:

Ryobi Green - REC 709 SAT 100

Ryobi-REC709-Sat_100.jpg (512.32 KiB) Viewed 14120 times

BUT - When you switch to YRGB Color Managed and REC 2020 in the project settings the color is much more accurate to begin with when you set the Saturation to 100 and still view it on a REC 709 monitor:

Ryobi Green - REC 2020 SAT 100

Ryobi-REC2020-Sat_100.jpg (496.73 KiB) Viewed 14120 times

When you switch to YRGB Color Managed and REC 2020 in the project settings the color is much more accurate to begin with when you set the Saturation to 100 and still view it on a REC 709 monitor:

Ryobi-REC2020-Sat_100.jpg

If you have start off with saturation at 100, doesn't that in itself point out there is a problem at the base level? And that Euginua is correct, it is something that needs to be addressed by BM?
Wouldn't a preference setting for their own cameras, that starts you with a built in correction for the way THEIR cameras deal with color, be the proper solution?

If I was getting a VO from a, Studio where I constantly had to boost the bottom end 100%, just to make the voice have some body, I'd be talking to the studio about checking their recording chain.

BUT - When you switch to YRGB Color Managed and REC 2020 in the project settings the color is much more accurate to begin with when you set the Saturation to 100 and still view it on a REC 709 monitor:

It sounds like you're what you're doing is super-saturating non-normalized log. If so, this is, to say the least, "problematical". You'll never get accurate colors that way. And viewing a rec. 2020 color space on a rec. 709 monitor raises a host of other questions.

Using the Ryobi shot you linked to, you can get fairly accurate color (at least, judging by the third-party product shots posted a few pages back) using the standard BMD Film-rec. 709 normalization. If you want more accurate color, apply a hue adjustment. Where it does actually turn yellow is when raising the highlights, as the OP apparently did, beyond the point of no return. But you'll get the same thing if you transform to Arri LogC and use Arri's normalizing LUT. So, if anything, it looks like it's the sensor, and that particular color.

It sounds like you're what you're doing is super-saturating non-normalized log. If so, this is, to say the least, "problematical". You'll never get accurate colors that way.

Using the Ryobi shot you linked to, you can get fairly accurate color (at least, judging by the third-party product shots posted a few pages back) using the standard BMD Film-rec. 709 normalization. If you want more accurate color, apply a hue adjustment. Where it does turn actually turn yellow is raising the highlights, as the OP apparently did, beyond the point of no return. But you'll get the same thing if you transform to Arri LogC and use Arri's normalizing LUT. So, if anything, it looks like it's the sensor, and that particular color.

Well. . . I showed that before in my first post with the Ryobi raw power washer clip link, that I could "normalize" it with a hue adjustment and kept the saturation down to 80%. So I don't disagree with you.

If you use REC 2020 for your settings you can immediately get accurate color as you saturate it without need for a hue adjustment. I'm not going to speculate about what is going on under the hood, but if you assigned me a product video where I had to work with the Ryobi green I might be inclined to use REC 2020 color space as Eugenia suggested, but I don't know if I need to go as far as using a cst to Arri.

I don't mind anyone downloading the 1gb RAW clip and checking it out for themselves, I used a Nikon 50mm 1.4G with a non speed boosting adapter. Let me know if I need to relink it.

One thing this thread has shown me, is to look at the various types of color options in the settings.

Using color management option is fine, however the timeline color space of rec2020 is factually too narrow for certain color ranges of BMD's pocket camera color science using color management in its current state, which means that you will need to adapt to this if you plan to use RCM with the pocket camera and want zero problems in the future. Its not a debate, you will have clipping in certain ranges and situations using rec2020 timeline.

Reds and blues for example can clip in rec2020 timeline but not as clearly as rec709 timeline. These values will not clip in Red wide gamut or the native color gamut of the camera along with some other gamuts you can choose from. You will not always see this clipping, but it does occur in certain situations like red/blue flowers and neon lights that appear as if they may be clipped in source but aren't.

The result of a rec709 conversion using either alternative to rec2020 will provide a similar rec709 output and any discrepencies, usually in green and blue, can be adjusted with hue v hue.

With a straight to rec709 color space transform and proper luma/saturation mapping, you can achieve accurate colors with a little bit of hue v hue adjustments provided the source had proper IR filtration.

I am not going to debate on how good or bad BMD's conversions are V other cameras, because honestly there's not much you can do to control BMD other than provide some comments.

But you CAN in fact control your own footage and correct color values to be exactly what you want with the tools provided.

Oh, man, Dmitry. This same youtubber -- again????

The BMPCC 4k/S1H comparison, with his unaccountable BMPCC 4K results, and the previous Jack dispute, who also relied on him, didn't discourage you?

I think maybe we need more reliable testing sources, before drawing conclusions about much of anything....

By the way, look closer at Youtube thumbnail. It seems that JSFILMZ always use ND filters. So yes, in addition to low quality grading his side by side tests (including test with tree and lake we discussed earlier here) are totally unfair and damaged by IR pollution.
Those tests only useful to compare how different cameras resist against ND filters IR pollution.

...I wish someone from BMD would comment.

I second this wish.

Steve

Here's screenshots with all my tests ( https://imgur.com/a/vzcaETl ), showing the problem, and how the linear + Arri is the best solution to make things look accurate. Sigma 17-50mm with Viltrox speedbooster, at f/2.0. Timeline was standard rec709, gamma 2.4.

Shot 1:
Normal rec709 with BDM colorspace and BDM Film on the raw tab, with BDM Extended lut. That's the standard, traditional grade. As you can see, everything has shifted to yellow-brown (yellow under full sun, yellow-brown in cloudy situations).



Shot 2:
The same as above, but having changed the colorspace to Rec2020 in the raw tab. Suddenly, the green is coming back. The colors on his t-shirts are also now more accurate! Not just the greens got fixed, but also his blues! I know, I have the t-shirt in front of me. Skintones are now MUCH better too, having shifted to the filmic/alexa yellow-ish color, rather than reddish colors that video cameras have. However, we're still not there.



Shot 3:
Melara's approach: Colorspace in the raw tab: Blackmagic design, Gamma linear. Then two color space nodes to arri alexa/logc, and then to rec709. One node in the middle to fix the contrast/exposure (because it's different when having transformed the image to Arri), so we can judge it better compared to the others. As you can see here, using Melara's approach, everything is as it's supposed to be. The greens are perfectly green, the other colors are also correct and skintones are pleasing. The sky now has color! THIS is the best result of all.



Shot 4:
The same as above, but instead of using Colorspace "blackmagic design" in the raw tab, I used rec2020. The colors did not change compared to Shot 3. Which means that BMD's colorspace is very close to rec2020 in reality (which is 75% bigger than rec709). But by going through the ARRI remapping to get to rec709 (since Resolve is probably using ARRI's software libraries to do so, instead of using its own algorithm), we get the proper colors.



Shot 5:
The same as Shot 1, but with the popular Butterly Lut instead. At first glance, it looks as if the lut has fixed the yellow cast. But in closer inspection, it has shifted the colors to cooler blue. Now the skintones and house paint are blue-er than they should be. Adjusting its WB by +300 points in the RAW tab, it again turns the greens towards brown. So these "fixing" luts seem to just shift the problem, not fixing it. Still, the Buttery lut gives a much better result than the hideous, default BMD lut.



Conclusion: You can try treating the footage as rec2020, which will fix quite a few things, but you will fall short of the Resolve rec709 remapping at the end. The most pleasing and accurate result, is the Linear->Arri->rec709 #3 method, because we bypass Resolve's internal algorithm for remapping, and we (probably) use Arri's. The image now has "life" in it. You *might* be able to fix things with regular color grading, but why have to go through hoops? Using Melara's approach, the colors are correct off the bat.

This is something that Blackmagic can fix in Resolve btw. They need to rethink their remapping techniques. No need to change any color science on the cameras themselves. The problem is in Resolve.

Rather than rehash these arguments, I have an experiment for anyone still on this mail run. Let's take Alexa LogC footage, which we'll normalize inside Resolve with the Arri-provided LogC to rec. 709 conversion LUT. As best I understand Eugenia's position, this procedure should provide an excellent result, since Arri, not BMD, wrote the LUT. The first Alexa clip on this page,

https://arriwebgate.com/en/directlink/51bfeb7a32fa6e8a

should do. It has challenging available light conditions, skin tone, bad white/yellow background wall, so a good case study....

Then we take the same shot and transform it to BMD log, using a CST OFX. The Transform fields will read, Arri Alexa, Arri LogC, BMD Gen4 Pocket 4K, BMD Gen 4 Pocket 4k. No mapping. Then we normalize this converted clip on the following node with the BMD Pocket 4K to Extended Video Lut.
In all cases, the CST and normalizing LUTs are on the latter nodes -- any basic corrections are done on prior nodes. And absolutely NO fooling with rec. 2020.

If Eugenia is right, the converted to BMD log shot should look like crap, because although it's Alexa, it's being transformed into BMD log and normalized with BMD math. And the straight Arri normalization should look great, or at least much better than the BMD version, because it's been normalized by the Arri rec. 709 LUT, not BMD math.

If anyone wants to point out methodological errors in designing this test, or my claims for what it should prove, ears are open. I'm not an engineer. But it seems to me a valid way to test the claims being made -- Alexa footage, with BMD math.

I could offer the results, but posting frames here appears to be 100% ineffective in changing anyone's mind, clear though I think the proofs are. Might as well do hand-stands. So try it yourself.

It has challenging available light conditions, skin tone, bad white/yellow background wall, so a good case study....

There is not much green in that footage, which is the main topic here. But even without much green, the shortcomings are visible. There is much less color information retained with the BMD container after the transformation.

Also, it's not the lut that makes the transformation properly. It's Arri libraries under the hood, possibly from an Arri SDK that Resolve uses when doing color transformations.

So, you can see the Arri footage with its lut, working as expected (even on a rec709 timeline and not in a managed Arri one -- which would be important for the reason mentioned on the paragraph above). Lush color, beautiful detail.



And here's the transformed BMD version with the BMD p4k extended lut, having lost both color and highlight information. You can see some emerald color on the back if you zoom in, that has desaturated and shifted colors. But in that footage specifically, the biggest problem is the loss of highlight and detail retention. If you look the intricate designs on the very back, they have lost saturation and were lost. Adjusting contrast/exposure did not bring them back.

Melara shows this very problem more clearly on his video, where the sun rays were not held well in the BMD format, but it was held up much better in Arri's.

I thought I dealt with the green earlier; the objections you raised to the shot I posted are, in my view, idiosyncratic. It's not pride in the work-- I did very little with the shot in post. But I think it's all working as expected. Bright unfiltered sunlight and high saturation; this is what happens. If that's not what's wanted, change the grade. It's easy enough to do.

In any event, my results with this Alexa shot are a bit different. The versions are barely distinguishable after fairly minor adjustments to the BMD version. You do have to grade a bit - it *is* different math, you wouldn't expect to get the same initial results with both. And the transformation is likely not without untoward consequences. The footage is not meant to be handled in this way. It would be surprising if the results were actually identical.

That aside, I can't assess your technical claims and explanations. I'm not an engineer and am not familiar with the programming involved. The source of your own knowledge of the inner workings here is unclear.

Anyway, none of this really matters. Your approach isn't one I would choose to take for myself, based on the results you've presented here. I don't see the claimed greater integrity of the footage you've produced; on the contrary, I think the footage is being corrupted, compared to the results achievable with recommended procedures. I should have let the whole thing drop, we're not going to get anywhere on this one, irreconcilable differences.

Even with my non-experienced eye, I definitely see the loss in detail. It's actually quite apparent on the face on the wall on the right.

The shots have to be graded to match. Or rather, the BMD shot has to be graded to match the Alexa shot, since that's the exercise. BMD normalizes footage differently than Alexa does. The match won't happen by itself.

And again, you wouldn't expect a perfect result with this unnatural process, which is another reason I'm skeptical of the transforms that Eugenia is performing. In retrospect, I don't think this was a great experiment. Too many uncontrolled variables, and not knowing what goes on under the hood.

The highlight color retention issue Eugenia attributes to Melara was later qualified by him, and effectively withdrawn. You can see his quote above, a few pages back.

The shots have to be graded to match. Or rather, the BMD shot has to be graded to match the Alexa shot, since that's the exercise.

BMD normalizes footage differently than Alexa does. The only way to judge this one is to do the work yourself. The match won't happen by itself. And again, you wouldn't expect a perfect result with this unnatural process, which is another reason I'm skeptical of the transforms that Eugenia is performing.

Ok, but why would I use a starting point that already has an issue?
And don't you think that BM actually needs to have a proper preset that incorporates the fix already?

Whether BMD math has an "issue" is the meat of the dispute here. Eugenia says yes, others of us are saying no.

The BMD normalizing LUT does a bit less initial work than the Alexa LUT -- the Alexa always looks better, when it's first applied. BMD made other choices, and they had their reasons.

Anyway, as I said, this isn't a good experiment anyway. Too many unknown variables, and most people here won't be inclined to work with the footage, without which the exercise is even less useful.

John, I followed your instructions, you didn't mention any matching on your first post about it. Besides, there is no way to match them for the things I mentioned (detail and highlight loss), because these are lost if using the BMD lut. Gone forever. While when you do the opposite, processing BMD footage via Arri's pipeline, then you GAIN information (information that was always there, but the default processing chain was not mapping well).

I thought it was self-evident, that the math of the two LUTs wouldn't be identical, any more than the normalizing LUTs are identical on their actual respective cameras, with the same footage. Remember Wolfcrow, claiming that all normalized footage should be identical, without any grading interventions?

Anyway, this detour wasn't helpful, it's just going to confuse the issue, proving nothing, one way or the other.

I'm the one insisting that the transforms can have unintended consequences, so it would have made more sense not to suggest one.

I actually think your test makes it crystal clear.
There needs to be a color space that fixes the problem for their own cameras. .

You can't draw that conclusion from what Eugenia posted, or from the test generally. You just can't. If you accept nothing else here I've said, please accept that.

I could post my own matched shots, which are much closer than Eugenia's (I graded that way), but this test is too poorly conceived to want to proceed with it.

Here's screenshots with all my tests ( https://imgur.com/a/vzcaETl ), showing the problem, and how the linear + Arri is the best solution to make things look accurate. Sigma 17-50mm with Viltrox speedbooster, at f/2.0. Timeline was standard rec709, gamma 2.4.

I can see relative differences but non of them 'look accurate'.

OP here.

Wow. I cannot believe what this thread has turned into. This is like pre-k for colorists. Bunch of 5 year olds running around screaming at each other and all the boys ganging up on Eugenia who seems to be one of the few here who is trying to come up with a PRACTICAL solution to the problem that makes sense to the average user of the BMP4K, NOT the highly experienced/trained colorist/cinematographer. If you read my first few posts, I made it very clear - I am not a pro, not trying to be one. All I wanted to do was find a simple solution to an obvious problem that I was having. To the ones saying that they can't see the problem - well that's just ridiculous. It's quite obvious from the image comparisons I posted. Please don't go into things such as IR pollution and a bunch of other terms that I have no idea what they mean. This was not some highly complicated setup. Just a 5500K LED light, a white piece of paper, and a green product on top of it. There was no pollution that could have caused such an extreme shift in the colors. To the ones constantly saying how "that workflow is so wrong and it doesn't work" while providing NO practical solution to the problem - thank you. You're being super helpful. (insert sarcasm here).

Now let's go back to what the post was originally about. I'd like to point out that I posted in the cinematography forum and NOT the color grading one for a reason. I will ask you one question:

Forget that Resolve exists. Forget all your fancy terminology. Imagine that all you have is your camera. Turn your camera on, white balance it properly, point it at a green object (ANY green object) and tell me - do you see true green or yellow-y green? You see yellow-y green. Big time yellow-y greens. Period. Even when applying the BMD LUTs that come with the camera which are supposed to give you fully graded footage straight out of the camera don't even come close to what the eye is seeing. Hate all your want but the bitter truth is that this camera simply cannot produce proper looking footage straight out of the camera. I can already hear everyone yelling "No camera can do that!!" But most other cameras are at least getting decent usable results. The BMP4K is not even trying to. It is so far out there detached from reality that ALL footage from this camera MUST go through a color grading process that is NOT simple to understand if you are not familiar with color grading in DR - and if you don't see a problem with that then you need to wake up.

At the end of the day, I should be able to point my camera at a green object, white balance my camera, and get footage straight out of the camera where the green object looks green - should I choose to do so. I currently cannot do that. If anyone can tell me how I can achieve that with ONLY my camera, my lens, a 5500K LED light, and a gray card for white balancing and nothing else - I'm listening. That should be all that is required for a simple task such as taking a shot of a green object. If I cannot achieve something as simple as that using just those items - then it's time to accept the fact that this camera has a major flaw.

Hi Ivan. I noticed the Lutify.me BMD Wide Gamut - BMD Film Pocket 4K EI400 to Rec709.cube provides a better interpretation of green for you. If you add this LUT to the camera, and apply the LUT to the footage directly, you can achieve what you want "with only your camera". If you do not find the Lutify LUT to your liking, there are a few others I can reccomend, but the process is the same.

Hi Ivan. I noticed the Lutify.me BMD Wide Gamut - BMD Film Pocket 4K EI400 to Rec709.cube provides a better interpretation of green for you. If you add this LUT to the camera, and apply the LUT to the footage directly, you can achieve what you want "with only your camera". If you do not find the Lutify LUT to your liking, there are a few others I can reccomend, but the process is the same.

Which package contains that LUT? I can't find it on their website.

@Ivan It should be a part of all their packages, it's in the "Log Conversion LUTSs" folder. LeemingLUT is also good, but their included Rec.709 LUT forces you to overexpose the image.

Turn your camera on, white balance it properly, point it at a green object (ANY green object) and tell me - do you see true green or yellow-y green?

It might have something to do with Bezold–Brücke shift:

"B–B hue-shift occurs in everyday observation of object colors: consider a red book or a green cloth, in- or out-doors, in highlight and shadow. Brightly lit areas are yellower than shaded ones (appearing redder or greener). Such effects (produced by the viewers’ visual system) are also commonly seen in color photography/television."
https://www.sciencedirect.com/science/article/pii/S0042698999000851

The hue perception of Ryobi green changes with brightness, as can be seen in this photo from Ryobi:

https://www.ryobitools.com/outdoor/products/details/40v-powersource-300-watt-battery-inverter-tool-only