GH5 to BMMCC conversion.......attempt!

[kye]

I've posted this idea previously, but for those that don't know, this is about trying to make a GH5 look like a BlackMagic Micro Cinema Camera.

I have a few goals with this project:

• Learn more about colour grading
• Learn what makes the Micro have the reputation it has for image quality (and Mojo!)
• Learn what I can do to make my GH5 footage better

The project was born out of the backlash of the P4K and the "it's too sharp" "it looks too modern" "they don't make them like they used to" and at that time I asked if anyone could shoot some P4K and BMMCC / BMPCC footage that I could play with, but it didn't work out.  I began thinking about that, and I realised that one of the foundations of my film-making philosophy is to capture the best image I can but to try and shift as much work to post as I can, where there's time to optimise and really get the best out of the footage.  Anything you learn in post can be retrospectively applied to all your previously shot footage, which for me as the family historian, has significant value.  This project then became partly about proving my own hypothesis, that if you get a good enough image then you can do whatever you like with it. 

So, I bought a BMMCC, a cheap colour chart, a halogen flood light, an IR & UV cut filter, and here we are.

Attempt #1 was shooting random footage and after I realised I forgot the IR/UV filter that was a bust.  
Attempt #2 was with a colour chart I printed off the internet, and gave me an interesting feel for what the footage was doing, with the various colour shifts that were exposure dependent and other squirly things going on.  I learned some interesting things from that, especially that the Micro did a lot of the things that film-emulations do, like rotate hues into the warm/cool axis and saturate that axis a lot more than the other axes.
Attempt #3 was with the same internet colour chart but starting again knowing more about the footage, and I learned a bit more about the nasties going on.

Which brings us to attempt #4.

This attempt was with the real colour checker, my face for some office-worker skin tones, and a different approach.

BMMCC frame (graded with a WB, Colour Space Transform, and no other alterations):

GH5 frame (attempt #4) :

GH5 frame (graded via Colour Space Transform from Rec.2100 to Rec.709):

One thing to talk about is that the GH5 was shot in normal 4K 10-bit HLG mode, but the HLG mode doesn't actually correspond to Rec.2100 or Rec.2020, and there isn't a clear specification that I could find that it actually aligns to, so there's kind of no standard way to convert it to Rec.709 that I know of.  Certainly the above isn't the nicest colour grade, but it's for reference.

The approach is based around a workflow that goes:

• Colour Space Transform from Rec.2100 to BlackMagic Design Film Gen 1
• Simple curve to get the greyscale roughed in
• Main colour node, with a Hue vs Hue curve, Hue vs Sat curve, Hue vs Lum curve
• Fine-tune colour node with Hue vs Hue and Hue vs Lum
• Hue vs Hue and Hue vs Sat selectively applied to more saturated colours
• Lum vs Sat to desaturate blacks and whites
• Saturation boost on skintones
• Minor Gaussian Blur to match sharpness levels

This isn't the only frame I am matching across - I took exposures 2 stops above the ones shown above, and another 2 stops below, and was matching those too.  The +2 image matches pretty well too, but I have more work to do on the -2, although unless you are seeing them side-by-side then it's probably difficult to tell.

The plan is to match:

1. Colour
2. Resolution
3. Mojo / Motion Cadence / Coffee-making-ability

I also shot the above frames in the 1080 10-bit ALL-I 200Mbps mode but haven't really looked at them yet.  I'll also likely include the 5K h265 mode too.

More to come.

[Deadcode]

Maybe i missed that part but why dont you shoot in vlog-l? Vlog to bmdfilm color space transform is a much easier starting point

[PannySVHS]

Thanks for sharing. Looks like a perfect match. Now, maybe at the end of your pipeline, could you put another curve and check if alterations look the same on both cameras? cheers

[kye]

10 hours ago, Deadcode said:

Maybe i missed that part but why dont you shoot in vlog-l? Vlog to bmdfilm color space transform is a much easier starting point

I don't own Vlog.  My plan is to give this conversion away when I'm done, so giving something away that requires the purchase of V-Log seems silly.

I'm also pretty sure that if I film in VLog then CST to BMD film the results will still require significant work, in which case what is the point?  If it doesn't require significant work then why aren't people just doing that in post, instead of moaning about how the GH5 looks too modern and the colour isn't great, if all it took was to buy a LUT and slap on a CST to get the colour of the 'baby Alexa'.

I see it like this...  the Micro has great colour, great image thickness, great resolution / sharpness, great mojo, but isn't great in battery life, sound, form factor, and file size.  the GH5 is great with battery life, sound, form factor, file size, but lacks the classic image.  

Think how foolish it would be if people have been complaining the whole time that the classic look has gone, when it was a LUT and CST away.  I doubt it.

4 hours ago, PannySVHS said:

Thanks for sharing. Looks like a perfect match. Now, maybe at the end of your pipeline, could you put another curve and check if alterations look the same on both cameras? cheers

Stage one is colour matching, stage two is resolution and image thickness, stage three is mojo.

By the time we get to stage three, we'll be doing blind comparisons of side-by-side 50p footage that's fully graded, and I'll be mixing up lenses between the two in order to eliminate what effects the glass is putting into the comparison.

I'm no-where near done  

[thebrothersthre3]

Well people have the Alexa lut from sage and people have called the OG pocket a baby Alexa. But people still aren’t satisfied. 

[kye]

16 minutes ago, thebrothersthre3 said:

Well people have the Alexa lut from sage and people have called the OG pocket a baby Alexa. But people still aren’t satisfied. 

People are never satisfied.  Mostly that's because they want something that they're not willing to put the work into getting.

I have the GHAlexa LUT from Sage and I'm still doing this.  I think having a LUT doesn't actually help a persons film-making all that much unless they were doing everything really well except colour, in which case that will significantly improve their weakest link.  Putting any LUT on uninspiring footage won't help at all, and buying a LUT without understanding what it's doing only gives you the choice of using it or not using it, which isn't actually that useful if you don't know how to colour grade yourself and can adjust shots to get the most out of the different lighting in each shot.

[kye]

Version 0.5

Have added in compensations for matching at low-levels and highlights.  These happened with some loss of fidelity in the mids, but we're probably well into nit-picking territory here.

I've added the ISO noise and someone in a PM said the Micro had more detail, so I've also removed the blur that I had on the GH5, so it's now a straight downscale to 1080.

I've also decided that it's now called GHMM.  If things don't have a cool name then google says they don't exist.

BMMCC +2 stops:

GHMM +2 stops:

BMMCC normal:

GHMM normal:

BMMCC -2 stops:

GHMM -2 stops:

There are a few hues where the brightness is quite a bit off, but these required very strong and very specific adjustments, which are likely to break real-world images, so I've kind of limited my grade to smoother adjustments.  There's still some riskier adjustments around the skin tones, but we'll see how well they go in real life in the future.

[kye]

This is the grade applied to the GH5 200Mbps 1080 mode - I'm interested to hear opinions on the resolution vs the 4K mode above.

GHMM 1080 normal exposure:

[kye]

If anyone wants a laugh, check this out.  

This is what happens when you put the GH5 and Micro side-by-side and film outside.  This is the exact same transformation as the above, which seems to match pretty well, but when you take it outside......

........not so much!!

Ha ha ha..  This is after setting the GH5 to WB of full sun, and carefully WB in post on both shots.  Other images are all screwed up by about the same amount.

The only variable here that is in play is that I had the Panasonic 14mm f2.5 lens on the Micro and the Panasonic 14-42 kit zoom on the GH5, so in theory there could be colour shifts, but both were set to F8 so were far away from whatever shifts might happen when they're wide open.

I guess if this stuff was easy then everyone would do it. lol.

[heart0less]

Wow, I wouldn't expect so drastic differences, but here they are..

Bummer.

 

Maybe that's why Sage offers 2 LUTs for his GHAlex - Daylight and Tungsten.

[Deadcode]

1 hour ago, kye said:

If anyone wants a laugh, check this out.  

This is what happens when you put the GH5 and Micro side-by-side and film outside.  This is the exact same transformation as the above, which seems to match pretty well, but when you take it outside......

........not so much!!

Ha ha ha..  This is after setting the GH5 to WB of full sun, and carefully WB in post on both shots.  Other images are all screwed up by about the same amount.

The only variable here that is in play is that I had the Panasonic 14mm f2.5 lens on the Micro and the Panasonic 14-42 kit zoom on the GH5, so in theory there could be colour shifts, but both were set to F8 so were far away from whatever shifts might happen when they're wide open.

I guess if this stuff was easy then everyone would do it. lol.

Omg, shocking!

I have tried to match sony colors to 5d raw acr default look with the method you tried. It took almost a year to realise: you cannot do this conversion with a single LUT for every light conditions

ACES conversion is the only way.

[kye]

Ah, ok..  I had forgotten that they were so different, so that makes sense.  

I can always build a different conversion for daylight.  I'd just need to film the requisite reference frames.

It seems odd that this would be the case though - essentially you just have a red sensor which receives light over an exposure and converts it to a digital value, and the same with green and blue.  If they happen to be inside or outside doesn't change this basic principle.  

I've been thinking that the way I processed v0.4 and v0.5 isn't the best way, as it doesn't involve a 3x3 or a YUV matrix, which is how I've seen lots of LUTs and film emulations reverse-engineered, and I would imagine is similar to how the Micro colour science acts too - it exhibits certain things that make me think that those things might be going on.  

Right - next attempt is to film some daylight reference shots and try to build a conversion that works across both.

[kye]

Juan just posted in his P6K to Alexa thread that he can do dawn to dusk with one transform, which makes sense to me.

I also disagree with the statement from @Deadcode that you can't do it in a single LUT, and that ACES is the way, because it makes me wonder what ACES is, if it's not a single transform?  and even if we take it that it's not a single transform, that would mean that the camera is changing the colour depending on circumstances, which to be properly processed would mean that the transform would have to know what colour the camera was using.

I just need more diverse data.  I have halogen, I can get natural light in a range of conditions, and I have the Aputure AL-M9 which claims a CRI of 95+, which will be good enough to use as a check, even if not as a source to match against. 

[Deadcode]

@kye do your homework, prove me im wrong

Sidenote: it will only work if the WB is set perfectly before shot and the exposure is the same in every shot. Just like with GHAlex.

[kye]

7 hours ago, Deadcode said:

@kye do your homework, prove me im wrong

Sidenote: it will only work if the WB is set perfectly before shot and the exposure is the same in every shot. Just like with GHAlex.

I realise that I might not have gotten the WB perfect on the outside shots - I set it to the outdoor preset instead of doing a custom WB, so I'll have to have a look at that.  I am curious about what effects getting the WB wrong actually has, apart from 'screwing it up' which we all knew.  I shoot auto-WB due to the situations that I shoot in so having a transform that is resilient to WB inaccuracies would be useful to me.  Of course, emulating the Micro and getting a decent transform are two separate things, but still.

Also remember that if I do it then I'll have proven you wrong, but if I fail to do it then you haven't been proven right, it's just that my attempt failed to prove you wrong  

[Deadcode]

7 hours ago, kye said:

I realise that I might not have gotten the WB perfect on the outside shots - I set it to the outdoor preset instead of doing a custom WB, so I'll have to have a look at that.  I am curious about what effects getting the WB wrong actually has, apart from 'screwing it up' which we all knew.  I shoot auto-WB due to the situations that I shoot in so having a transform that is resilient to WB inaccuracies would be useful to me.  Of course, emulating the Micro and getting a decent transform are two separate things, but still.

Also remember that if I do it then I'll have proven you wrong, but if I fail to do it then you haven't been proven right, it's just that my attempt failed to prove you wrong  

AWB is a no go.

When i tried to match my a6300 to my 5D i tried to measure the AWB differences. The problem is: In broad daylight the WB diff with the same lens was around 2-300K, indoors it was 5-600K or even more. The Sony went for the perfect WB, the Canon went for the moody WB...

I created WB shifter LUT's for my daylight/indoor LUT, and i always had to touch the image, it was never matched perfectly.

Would you please share some footage? im curious about the BMMCC image.

[Juan Melara]

It's not easy is it!

To make it easier, the best workflow is to linearise the GH5, make the corrections, then return it back to log. That means shooting in a gamma that can be converted to linear via a CST, or via something like Lattice.

If you don't linearise, you'll be forever chasing your tail. A correction at 0EXP will throw something out at 2EXP, fix it at 2EXP and you've just thrown it back out at 0EXP. When you linearise you remove the variability of the log curve, which can make it so much easier.

[kye]

6 hours ago, Deadcode said:

AWB is a no go.

When i tried to match my a6300 to my 5D i tried to measure the AWB differences. The problem is: In broad daylight the WB diff with the same lens was around 2-300K, indoors it was 5-600K or even more. The Sony went for the perfect WB, the Canon went for the moody WB...

I created WB shifter LUT's for my daylight/indoor LUT, and i always had to touch the image, it was never matched perfectly.

Would you please share some footage? im curious about the BMMCC image.

AWB is what I shoot, and I think it just depends on how close a match I really want.  The more difference I can tolerate the more flexible the inputs will be.

What kind of footage would you like?

18 minutes ago, Juan Melara said:

It's not easy is it!

To make it easier, the best workflow is to linearise the GH5, make the corrections, then return it back to log. That means shooting in a gamma that can be converted to linear via a CST, or via something like Lattice.

If you don't linearise, you'll be forever chasing your tail. A correction at 0EXP will throw something out at 2EXP, fix it at 2EXP and you've just thrown it back out at 0EXP. When you linearise you remove the variability of the log curve, which can make it so much easier.

Ha ha ha..  no, not easy at all.  But that's half the fun isn't it?  I don't like easy challenges  

Thanks for the tips, I'm basically doing this on what I've learned from messing around in Resolve and watching your LUT re-creation videos a million times!

Would you suggest literally doing the corrections in linear?  What tools would I use - my understanding is that the LGG wheels are in rec709 and offset is in log, but I don't know which would be that useful in linear?

[kye]

Last night I confirmed that the Panasonic 14mm f2.5 and the Panasonic 14-42mm kit lens are very similar in colour, so are close enough to be considered identical, which means I can shoot with both the GH5 and the Micro at the same time.

This is really important as it means that I can shoot clips for comparison in rapidly changing lighting, instead of having a gap where I have to move the lens between cameras and the light might change during that delay.

In order to accomplish this, I've just spent / wasted about an hour finding a rig that allows both cameras to be usable at the same time, where I can access everything on both, and can be tripod mounted or hand-held.. Behold the ugliness!

I've also worked out a bunch of lens combinations that will have the same FOV where the "good" lens is sometimes on the GH5 and sometimes on the Micro.

My plan is that once I've matched the colour and resolution we can get to testing mojo, and for that I want to be able to eliminate the factor of the lens, and I'll also be doing some blind (and anonymous) tests to see which people prefer, so having a rig that I can shoot with easily will be critical.  As such, I had to work out all the combinations of lens filter thread sizes and making sure that every lens on both could get a variable ND and the Micro would always have the IR /UV cut filter too, so all those are now ordered.  That was another hour of my life I'll never get back!

I've also been thinking about what Juan said and I think I've wrapped my head around it and now have a plan.... 😈😈😈

[kye]

I'm now more confused.

I think this is the system we're dealing with:

The challenge is that on the Micro I am shooting RAW and so the only colour variable is the RGB filters on the sensor, but with the GH5 the variables are (something like) the RGB filters on the sensor, the WB setting, and then the colour profile, which is likely to be a 3D LUT with all sorts of unknowns in there.

A realisation that's useful is that I can take a RAW still image on the GH5 and that should allow me to do a straight 3x3 adjustment in Linear to match them, without having to worry about the WB and 3D LUT that's on the video files.  I should also be able to lock the GH5 down on a tripod and take a series of RAW images and video files and then I can try and use the RAW image as a reference to isolate the WB+3DLUT away from the RGB filters on the sensor, although this doesn't seem to help that much as the 3D LUT is where all the nasties are.

I suspect that:

• the RGB filter adjustment is a 3x3 RGB mixer adjustment in Linear
• the WB adjustment is an adjustment of the levels of RGB in Linear
• and that the 3D LUT is all sorts of wonkiness

I've also worked out that I can use my monitor as a high DR and high saturation test-chart generator.  It won't be calibrated, but it shouldn't matter because the entire idea of this conversion is that any light that goes into either camera should result in (broadly) the same values after my conversion, so absolute calibration shouldn't matter.

I've also realised that this is a mixture of black-box reverse-engineering where there are a few black-boxes and we can take guesses about how each is likely to work.  Of course, Resolve isn't a black-box reverse-engineering tool, and it's very difficult to get pixel-perfect comparison images between the two cameras so that's kind of what makes it a challenge.  I'm wondering if I should be converting things in reverse order, peeling the onion at it were, so doing the opposite of the 3D LUT, then the WB, then the RGB filter.

I've also realised that I should switch to Cine-D profile instead of HLG, because although it gives a slight reduction in DR, it will enable the conversion to work on >30p footage too.