Inside BMPCC DNG & A Better Demosaic

Well, 3 exports yeah. edit: I yeah, I suppose it's LMMSE, DCB and VNG4 that's true

It's necessary to bring at least one TIFF sequence into resolve anyway, to conform with your edit, and then you can rebuild highlights and target whatever display, rec709 or P3 or what ever. Rawtherapee was never going t be where you get the final export. And for one pass, I would use LMMSE, since the zipper artifacts are less noticeable, and the noise performs better.

Adding DCB+VNG4 is an extra step which solves another two artifacts, the screen door and the zippering.

You can also let Raw Therapee load the white balance and color profile from the camera if you don't need highlight recovery or good color science in tungsten lighting, and then the 'false color suppression' tool on the raw tab will work, and you can reduce color artifacts on a DCB export instead if you just want one pass.

Thanks for sharing. I’m not sure if some of the artifacts annoy me enough to justify the extra time. But it’s always good to have an alternative at hand.

I also wanted to chime in and express my appreciation for this thread, as a heavy Micro user. The moire really is the most annoying thing with this camera.

I've been looking into the fixed pattern noise and it's a surprisingly difficult challenge to remove it. I think this is because this is a dual gain sensor, so that the fixed pattern noise and hot pixels actually show up 'twice' in the image, depending on exposure.

The sensor in the BMMCC/BMPCC works by reading each photosite twice at different exposures, and then adding them, and it must have some selection of which signal path, high or low gain, to use. So any fixed pattern noise removal has to be exposure sensitive.

The problem is if you subtract fixed pattern noise by recording frames with the lens cap on, it won't have much effect on where fixed pattern noise shows up in the image, and instead just clean up the absolute noise floor.

But if you subtract by recording the noise where it first shows up, by recording frames of an underexposed grey card, the deep shadows will have too much subtracted and it actually adds an inverse of the pattern to the lower shadows, which basically looks like an even worse FPN. Though I think this would be preferable even though it destroys more of the image overall, it pushes the fixed noise down further so that it won't ever show up at 800iso or maybe even 1600iso.

I did find Raw Therapee has a 'dark frame' folder and if you put in your fixed pattern noise frames to subtract it does indeed average them before subtraction even though it doesn't specifically say this, and it does it before the debayer. I also tried using a software called Siril which can give you a TIFF average of the raw frames, undebayerd, and renaming the TIFF to .DNG tricks rawtherapee into using it as a dark frame, no need to change the metadata. But the results were basically identical. You have to adjust the 'raw black points' in the raw tab after doing this or else your shadows will look crushed from the subtraction.

I also found putting the camera to 1600iso in raw and recording a grey card at just under 10IRE was a great spot to get the dark frames for subtraction. You just need two or three seconds worth, 50 - 100 frames is more than enough.

If you absolutely needed the noise floor, you could do two exports, one with subtraction frames generated with a grey card 1600iso at 10IRE and one with frames generated with the lens cap on, and then with an alpha channel blend the two exports. Resolve's controls are not fine enough to do this in linear space but you can transform into and out of something, and if you matched the black points between the two, it's extremely easy to get a clean result.

The problem is if you subtract fixed pattern noise by recording frames with the lens cap on, it won't have much effect on where fixed pattern noise shows up in the image, and instead just clean up the absolute noise floor.
But if you subtract by recording the noise where it first shows up, by recording frames of an underexposed grey card, the deep shadows will have too much subtracted and it actually adds an inverse of the pattern to the lower shadows, which basically looks like an even worse FPN.

Yep, i can confirm, came to exactly same results long time ago when attempt to subtract FPN viewtopic.php?f=21&t=65149#p536864
As i remember it was a detailed thread om bmcuser, but it is not available anymore.
It is really interesting theory that dual non uniform FPN caused by gain.

If apply TNR to original ISO800 source and only then increase Exposure (Gain in Linear gamma) for already de-noised footage - final image looks very clean, but FPN became way more visible from some point.

Did you use NeatVideo to subtract it?

I created single FPN frame in Photoshop from about 100 frames stacked in Mean or Median mode. Affinity Photo can do Mean or Median stacking as well and process way faster and easier than Photoshop.

I'm not sure if I should make a new thread or just keep all my tests in this thread, let me know if one or the other is better.

I noticed when I used my speed booster and a fast lens that I had what I thought was really bad longitudinal chromatic aberration (LoCA). I tried to see if anyone else noticed this but it seems speedboosters never caused this for anyone else. After I noticed it there, I realized it was showing up with all my lenses... speedbooster or MFT native... It was just more noticeable with shallow depth of field, and that it wasn't exactly LoCA.



So what it looks like is out of focus areas seem to be bluer at their edges, and it seemed worse in the left-right direction.



That's super weird, right? It's pushing the blue channel out of the center of the bokeh. If I rotate the camera, the left right spread stays in line with the frame.

The only thing I can think of is the Rawlite OLPF. One possibility is that since the olpf has a rectangular baffling, that's interfering with the bokeh. But I somehow expect it to be on shorter axis, up and down, and I can't picture why it doesn't effect the other channels at all, and it's only left and right.

So my best guess is that one layer of the bifringement material that makes the OLPF work is somehow much stronger on the blue channel and since it sits kinda far away from the image plane it has a good chance to mess up the bokeh. I don't want to take the olpf out to check since it was a little intense putting it in.

Can anyone who has a rawlite installed - especially if you have the bmpcc specific speedbooster - reproduce this weird effect?

I'm not sure if I should make a new thread or just keep all my tests in this thread, let me know if one or the other is better.

I noticed when I used my speed booster and a fast lens that I had what I thought was really bad longitudinal chromatic aberration (LoCA). I tried to see if anyone else noticed this but it seems speedboosters never caused this for anyone else. After I noticed it there, I realized it was showing up with all my lenses... speedbooster or MFT native... It was just more noticeable with shallow depth of field, and that it wasn't exactly LoCA.



So what it looks like is out of focus areas seem to be bluer at their edges, and it seemed worse in the left-right direction.



That's super weird, right? It's pushing the blue channel out of the center of the bokeh. If I rotate the camera, the left right spread stays in line with the frame.

The only thing I can think of is the Rawlite OLPF. One possibility is that since the olpf has a rectangular baffling, that's interfering with the bokeh. But I somehow expect it to be on shorter axis, up and down, and I can't picture why it doesn't effect the other channels at all, and it's only left and right.

So my best guess is that one layer of the bifringement material that makes the OLPF work is somehow much stronger on the blue channel and since it sits kinda far away from the image plane it has a good chance to mess up the bokeh. I don't want to take the olpf out to check since it was a little intense putting it in.

Can anyone who has a rawlite installed - especially if you have the bmpcc specific speedbooster - reproduce this weird effect?

What speedbooster do you use?

Metabones super 16 0.58x.

I'm seeing it on MFT lenses without using the speed booster as well. I think that along with the fact that it lines up with the frame narrows it down to the Rawlite OLPF or maybe the sensor. But it's hard to imagine how. I never saw it in other footage so I guess it's the rawlite - but why only the blue channel?

Speed boosters absolutely add aberrations.

It’s especially noticeable on faster lenses, say under F2.0

A lot of OEM MFT lenses tend to be less corrected because they assume relying on the camera to recognise which model lens it is and automatically apply corrections for many aberrations.

Which MFT lenses are you comparing? Is it the same issue when you use the speedbooster?

JB

The speedbooster is a red herring, it's on lenses even without it. I think it's not possible to be the lens because all the optics are circular, but this aberration is left-right only, aligned with the frame, and consistent between the couple lenses I tried.

Some sensors were designed in a way to have issues with wide lenses and short flange depth, but that's also not what this is, I don't think, and it shows up even on a 35mm.

So it has to be between the lens and the sensor, which is the rawlite OLPF I installed. It does sit a distance from the sensor and it's job is to turn one peak into two peaks (and then 2 into 4 peaks) like the blue channel shows - what should be a peak divided into two... But it's pretty weird.

Maybe the lenses have enough LoCA to send the blue channel into the OLPF sharp enough to get separated but the red and green focus later and get through unscathed? anyway probably no way to fix it...

edit: and just after checking, I thought to look at the foreground bokeh, they don't have it. they're just normal bokeh.

Try looking up sensor shading

JB

It's not that either, but thanks. If I stop down I can get a very flat illumination.

This is where the bokeh has pushed apart to the left and right on the blue channel, like a double image, rather than any kind of vignette.

edit: here's another way to see it. The cores of the blur of the bars here are purple.



I rotate the camera 90 degrees, same position. The cores are black like they should be, and also the blur is less spread out.



Have you ever seen out of focus blur that depends on orientation or how level the camera is? Maybe only with an anamorphic lens, which this isn't. it's super strange



I wonder if this can happen on a camera without the rawlite OLPF, or if it's the OLPF causing it and why? I don't see how it could be anything else.

It could be the Rawlite. You have one fitted?

If it’s not flat. Or has a manufacturing defect.

Can you swap it out to test out with another copy?

JB

Thanks for your help, I think I'll learn to live with it. I also thought it was maybe an alignment issue but it seems everything is flat and inserted properly.

I considered removing the OLPF and getting another gasket to put it back after testing but I decided even if it was from the olpf, the olpf corrects worse aberrations than it causes. If I removed it and it was still there I also don't know how to resolve it.

Ultimately, its very rare to focus within 1' wide open to get such large bokeh, so it's not a problem for most actual footage. thanks again.

Which version of Rawlite did you used? Old version in plastic enclosure had sort of internal reflection and purple/violet flare problem if bright light located somewhere above lens. Actually it didn't looked like flare, but more like color gradient. I wonder if same reflection could cause color cast in de-focused areas somehow.
I discovered and reported this bug and later problem was fixed in new version of Rawlite with aluminum CNC machined enclosure.

I still have Mosaic instaleld on one camera and Rawlite on another, so i probably could compare.

I have a recent version, purchased from rawlite in the last 6 months, but it felt like plastic so maybe old stock. But, I no longer think it's the OLPF.

I now think it is possibly a kind of sensor shading, something to do with the microlenses on the sensor. I was mistaken about ruling that out.

I coincidentally saw this on vimeo about an hour ago and thought it looked interesting, I never heard of this product and 8 years ago I'd never even heard of aputure, interesting that they were making things like this, but then, at 1:36, the bokeh with this speedbooster has the same left-right blue separation, identical. I even took a screenshot and took it into resolve and the waveform has the same dip. It's happening with orange lights, just like how it first appeared with my issue. This is on the pocket OG, but same sensor and optical path so that shouldn't matter.



If it really is something to do with the sensor my only remaining question is if it's only with certain lenses.

I guess the easiest way to test is to take a lens with good close focus and larger aperture and record the bokeh of an orange light, or do the rotation trick with horizontal/vertical lines like my other post shows.

Is it changing colour as you rack focus?

JB

Which version of Rawlite did you used? Old version in plastic enclosure had sort of internal reflection and purple/violet flare problem if bright light located somewhere above lens. Actually it didn't looked like flare, but more like color gradient. I wonder if same reflection could cause color cast in de-focused areas somehow.
I discovered and reported this bug and later problem was fixed in new version of Rawlite with aluminum CNC machined enclosure.

I still have Mosaic instaleld on one camera and Rawlite on another, so i probably could compare.

Just to clarify, when did this change occur and when was the new version manufactured?

I know there is a V2 of the Rawlite OLPF for the BMMCC, I had actually made a post inquiring about the differences about a month ago.

Is that what you are referring to? Is the V2 worth “upgrading to” then?

Just to clarify, when did this change occur and when was the new version manufactured?

From my email conversation archive, new version was available somewhere from the end of 2020 / beginning of 2021. Here is how they both look:

IMG_7788.jpg (603.4 KiB) Viewed 2767 times

Oh this is great, and based on this o think actually already have the new version so thanks so much for sharing!