Denoising diminishes braw and prores detail

And here I placed all clips into a HD 1080p timeline with no scaling and center crop:

CDNG uncompressed

HD Centercrop_1.1.1.jpg (814.65 KiB) Viewed 15237 times

BRAW Q0

HD Centercrop_1.2.1.jpg (706.35 KiB) Viewed 15237 times

ProRes HQ 422

HD Centercrop_1.4.1.jpg (672.32 KiB) Viewed 15237 times

And here I placed all clips into a HD 1080p timeline with no scaling and center crop:

HD Centercrop_1.1.1.jpg

HD Centercrop_1.2.1.jpg

HD Centercrop_1.4.1.jpg

Hey man thanks a lot for doing this little test for us. Granted it's not the ideal condition, with higher contrast lighting and textures. But still thanks a lot for this man, beautiful city you live in.

I will upload the original footage to WeTransfer when I am back in office later.

I will upload the original footage to WeTransfer when I am back in office later.

Thank you very much for this.

Here are the direct PNG exports from Resolve:

https://we.tl/t-QQv9EdsRf0

Here are the direct PNG exports from Resolve:

https://we.tl/t-QQv9EdsRf0

Thank you very much for this Robert, you rock.
In this scenario braw is actually doing great, I would have no problem cropping it 1.5x in these particular high contrast situations with high contrast textures. If anyone can somehow shoot some low contrast situations with some relatively far away faces (like the jsfilmz samples) that would be very nice. I personally think this sample demonstrates that the braw should have decent resolution when the denoiser isn't smearing out some low contrast textures.

It's not necessarily just denoising at all. One of the principles of DCT compression is loss of detail where human vision doesn't expect it. We don't see small detail in low-contrast areas too.

Imagine a wall with fine grain painted white. You will simply see it as white when the light is soft. But if directional light is hitting at a flat angle, you'll see every little piece of grain from the same distance.

Do a simple test is not hard, do a good shooting and press photo button, and do simple “check the difference”.

http://www.macchiavello.com/wp/en/shoot ... a-cameras/

Here you can find an article that I wrote last years like how to shoot photo with cinema camera, where I shoot uncompressed cdng against extracted frame from Q0. Check the picture and tell me if you find difference between extracted frame and uncompressed cdng. I usually use a photo camera, but my photo camera had an accident night before then I use that I have.

If you want to do a test you must do at best of light and picture when you can see the sharpness, if you talk about sharpness.

And remember that you never shoot at cdng at that quality, be cause this is uncompressed and shooting was (my pocket are from original batch) 3:1 or more compressed.
Cdng was good in 2012 with some sensors but today I find that codec like braw is the best combination of quality - go - cpu/gpu stress.

Me I had gh2, I was in the group of hacking mod of gh2, never had the sharpness of pocket, also if shoot at 12:1.. last week when I tried to cut some old shooting with newer seems that I leave my eyes glasses on table when I saw gh2 shoot.


Inviato dal mio iPhone utilizzando Tapatalk

I have left some samples in my original post that demonstrated what I was talking about. And when I mentioned the gh2, I was talking about hoping that braw would match the gh2 in 1080p while being cropped to 1.5x. Braw is fine if you don't try crop. But for people like us who needs to crop a lot the softness of braw gets greatly exaggerated. I don't need cdng, I want braw with user selectable level of denoising. Also the download on your website seems to be failing after a few megabytes.

Two different thread, if you want crop later to emulate two cameras you can’t think that a 4K down samples to 1080 could have same quality of cropped 1080p.

Gh2 has the same ability, shoot down samples 1080p from 16mpx sensor or with special zoom hardware you crop sensor to 1080p, which be most of time useless due noise if you not shoot in native iso and with a lots of light.

Also at best gh2, with modified firmware to shoot at 200mbt, modified matrix compression is not so sharp like a 4K down sampled to fhd

Anyway denoiser work in p4k different between different iso / light presence setup, that mean is not easy to setup on camera.
Work different from motion to static picture, then if you had different subject different setup could give you more or less then you want.

I tried myself to download file from my website and work fine, try again


Inviato dal mio iPhone utilizzando Tapatalk

Here are the direct PNG exports from Resolve:

https://we.tl/t-QQv9EdsRf0

Thank you very much for this Robert, you rock.
In this scenario braw is actually doing great, I would have no problem cropping it 1.5x in these particular high contrast situations with high contrast textures. If anyone can somehow shoot some low contrast situations with some relatively far away faces (like the jsfilmz samples) that would be very nice. I personally think this sample demonstrates that the braw should have decent resolution when the denoiser isn't smearing out some low contrast textures.

There's plenty of low contrast areas in those images - low contrast is not determined by lighting alone. As I doubt that BM has any AI denoising that is selectively applied to faces I don't think you are going to get a better test than this. Maybe another test is to compare under exposed images to see how much BRAW is applying NR to the lower parts of the data range but these tests pretty much mirror my own expectations and experience of BRAW and CDNG.

Do a simple test is not hard, do a good shooting and press photo button, and do simple “check the difference”.

http://www.macchiavello.com/wp/en/shoot ... a-cameras/

Here you can find an article that I wrote last years like how to shoot photo with cinema camera, where I shoot uncompressed cdng against extracted frame from Q0. Check the picture and tell me if you find difference between extracted frame and uncompressed cdng. I usually use a photo camera, but my photo camera had an accident night before then I use that I have.

If you want to do a test you must do at best of light and picture when you can see the sharpness, if you talk about sharpness.

And remember that you never shoot at cdng at that quality, be cause this is uncompressed and shooting was (my pocket are from original batch) 3:1 or more compressed.
Cdng was good in 2012 with some sensors but today I find that codec like braw is the best combination of quality - go - cpu/gpu stress.

Me I had gh2, I was in the group of hacking mod of gh2, never had the sharpness of pocket, also if shoot at 12:1.. last week when I tried to cut some old shooting with newer seems that I leave my eyes glasses on table when I saw gh2 shoot.


Inviato dal mio iPhone utilizzando Tapatalk

I have left some samples in my original post that demonstrated what I was talking about. And when I mentioned the gh2, I was talking about hoping that braw would match the gh2 in 1080p while being cropped to 1.5x. Braw is fine if you don't try crop. But for people like us who needs to crop a lot the softness of braw gets greatly exaggerated. I don't need cdng, I want braw with user selectable level of denoising. Also the download on your website seems to be failing after a few megabytes.

Two different thread, if you want crop later to emulate two cameras you can’t think that a 4K down samples to 1080 could have same quality of cropped 1080p.

Gh2 has the same ability, shoot down samples 1080p from 16mpx sensor or with special zoom hardware you crop sensor to 1080p, which be most of time useless due noise if you not shoot in native iso and with a lots of light.

Also at best gh2, with modified firmware to shoot at 200mbt, modified matrix compression is not so sharp like a 4K down sampled to fhd

Anyway denoiser work in p4k different between different iso / light presence setup, that mean is not easy to setup on camera.
Work different from motion to static picture, then if you had different subject different setup could give you more or less then you want.

I tried myself to download file from my website and work fine, try again


Inviato dal mio iPhone utilizzando Tapatalk

Oh I meant hoping the bmpcc4k would match the uncropped full sensor hd of the gh2 while the bmpcc4k is cropping in 1.5x. The gh2 pixel bins while shooting full sensor hd, so that's why I think that it's a valid comparison. And yeah the dnenoiser works different in every lighting setup and iso, so I think if there were like "very low, low, mid, high" That would be good enough. Though the download is still not working

Yes, I was going to say a few more things Jack, but by the time I got through it. I was going to ask if you were using a lower datarate modes instead of the highest one. That would damage the footage, despite people around stating they can not see a difference between lower modes and 12:1. There was something else important, but I just forgot it again. Anyway, what you are doing requires the best mode to get extra quality. Things get sacrificed at the very small levels, so doubling the size of those makes things more apparent.

Yep, I remember what I was going to say in a previous post. It maybe the noise reduction is part of the heavy quantitisation, thus intrinsic like John implies. The clow the quantisation takes things to no difference to the underlying image curve, the more "muddy" it seems. It's a thing we get in consumer video too
Than there is extensive extra wide demosaicing and anti-alaising, all which the above curves can help archieve, especially if the curves are dynamic in length, or overlap. However, cdng lowest compression mode and highest data rates prores, show better can, and was, done on this camera in the past,cabs it's built for it.

Sorry for the memory lapses, with all the freaky stuff happening here, I'm not physically keeping up with the treatments for brain disease causing the memory issues. However, a few things have put me on treating some previously unknown stuff, improving things, which I have to check if it caused insect damage mistaken for the brain disease. If right, I can hopefully stop more degradation.

Here are the direct PNG exports from Resolve:

https://we.tl/t-QQv9EdsRf0

Thank you very much for this Robert, you rock.
In this scenario braw is actually doing great, I would have no problem cropping it 1.5x in these particular high contrast situations with high contrast textures. If anyone can somehow shoot some low contrast situations with some relatively far away faces (like the jsfilmz samples) that would be very nice. I personally think this sample demonstrates that the braw should have decent resolution when the denoiser isn't smearing out some low contrast textures.

There's plenty of low contrast areas in those images - low contrast is not determined by lighting alone. As I doubt that BM has any AI denoising that is selectively applied to faces I don't think you are going to get a better test than this. Maybe another test is to compare under exposed images to see how much BRAW is applying NR to the lower parts of the data range but these tests pretty much mirror my own expectations and experience of BRAW and CDNG.

Oh yeah that could be a good test as well, but what I mean by low contrast texture is like same colored fabric. Where with cdng you can see some threads while on braw it's just a smooth surface when testing at a particular distance. And at least for me showing far away faces is usually the most obvious places that you can see the smearieness, the eyebrow and nose just become one.

Yes, I was going to say a few more things Jack, but by the time I got through it. I was going to ask if you were using a lower datarate modes instead of the highest one. That would damage the footage, despite people around stating they can not see a difference between lower modes and 12:1. There was something else important, but I just forgot it again. Anyway, what you are doing requires the best mode to get extra quality. Things get sacrificed at the very small levels, so doubling the size of those makes things more apparent.

Yep, I remember what I was going to say in a previous post. It maybe the noise reduction is part of the heavy quantitisation, thus intrinsic like John implies. The clow the quantisation takes things to no difference to the underlying image curve, the more "muddy" it seems. It's a thing we get in consumer video too
Than there is extensive extra wide demosaicing and anti-alaising, all which the above curves can help archieve, especially if the curves are dynamic in length, or overlap. However, cdng lowest compression mode and highest data rates prores, show better can, and was, done on this camera in the past,cabs it's built for it.

Sorry for the memory lapses, with all the freaky stuff happening here, I'm not physically keeping up with the treatments for brain disease causing the memory issues. However, a few things have put me on treating some previously unknown stuff, improving things, which I have to check if it caused insect damage mistaken for the brain disease. If right, I can hopefully stop more degradation.

I see, I did shoot at q0 and 3:1 for max datarate for my samples, and the jsfilmz samples were also shot at q0 for max data, so I think they represent what braw can achieve with the most amount of data. It does seem like the denoising is more complicated then I first imagined it to be. I just hope that blackmagic can somehow improve this problem in a future update. Best wishes to your health as well mate,

Thanks, I see. The reality is Braw might need 1.5:1-2:1 to get the most. The 3:1 is probably 4:1+ on an non bayer internal component video format, that's an issue. At 2:1 it probably will be closed to 3:1 internally. I'm not looking for something from nothing.

I'm scrolling around the BRAW vs CDNG in Photoshop at 100% on a UHD monitor (not far off my nose) There is plenty of low contrast detail if you look hard enough esp in shadow areas. What I can see is lots more false detail and aliasing in very fine detail in CDNG. Very little if any difference in low contrast areas. Some difference in medium detail like foliage . My overall take is the general level of aliasing and false detail is more obvious and distracting than any advantage in apparent detail in other areas and I'd be wanting to apply something on post to knock this false detail back. I'll look at the footage when it's available as movement is another factor in how aliasing looks which can't be 100% established in a frame grab. Case closed as far as I'm concerned. Nothing here that makes me change my initial opinion and can't see how shots of distant faces or fabric are going to change that for Jake (maybe shoots with some straw maybe useful to clutch at )
One other important observation is that at the frame edges where the lens resolution drops off the differences are very much reduced. This reinforces my view that critical focusing and top quality glass ( including ND filters as Robert has used) are not optional if you want to pull the best off this (or any sensor) and especially so if you are wanting to punch in.

And here I placed all clips into a HD 1080p timeline with no scaling and center crop:

HD Centercrop_1.1.1.jpg

HD Centercrop_1.2.1.jpg

HD Centercrop_1.4.1.jpg

Robert, thanks for this. A shame many who want to argue against things will not put the effort in to test prove things.

I have looked at them now. Guess which one looked best when I first saw it at 2 inches or less in my phone? The cdng. Amazing, but braw dues take something away frombth image. If you look for actual missing details, it may not find much off of surfaces fairly flat, but it is about the levels of neighbouring pixels being squashed so detail seems to fade. Also remember the screen with mesh, where the mesh strandels just became a semi transparent block, instead of individual strands with gaps between to the scene behind. I think that was at the same time as the bank note and Supernan figurine tests samples. Strange to look at. But if BM wants to be really professional they needs more options like what we are taking about for extra quality for case scenarios, like Jack's.

Firstly, thanks Robert for re-doing the test that Jack should have done properly in the first place.

With your well documented methodologies if I was doing this I would have dropped the VND as well because they are known to alter fine detail (even the good ones)

I am currently locked down in New Orleans have don’t have any BMD cameras with me, only stills cameras or I would have also done this.

Again though the best way to test jacks claim is to replicate it with faces. The onus primarily being on him to do it because then it’s with the variables he has and uses.

Using other people’s footage is less useful. Jacks already I think said he would “pass” Roberts test with qualification. It’s bright and no faces ?

So again in circles.

I really think a lot of this would have ended with the right examples in the first place.

JB

One person's vital information that needs to be retained is another's unwanted noise that needs to be filtered out. As it is with sensors and codecs so it is with forum threads discussing the same.
Noise/artefacts/image detail (take your pick) in the CNDG only equals one thing we can all agree on which is more data to compress, transfer and store.

Firstly, thanks Robert for re-doing the test that Jack should have done properly in the first place.

With your well documented methodologies if I was doing this I would have dropped the VND as well because they are known to alter fine detail (even the good ones)

I am currently locked down in New Orleans have don’t have any BMD cameras with me, only stills cameras or I would have also done this.

Again though the best way to test jacks claim is to replicate it with faces. The onus primarily being on him to do it because then it’s with the variables he has and uses.

Using other people’s footage is less useful. Jacks already I think said he would “pass” Roberts test with qualification. It’s bright and no faces ?

So again in circles.

I really think a lot of this would have ended with the right examples in the first place.

JB

I am also in a lockdown, with all of my previous tests being done in a production, so unfortunately I can not share it. And of course that's on me for not providing enough examples, but the jsfilmz sample basically demonstrates what I was talking about when I talked about muddiness on the far away faces and low contrast textures (https://drive.google.com/file/d/1HjxhVP ... pYS8M/view). And unfortunately the sigma 17-50 is all I got with me in this lockdown, but the low contrast smearieness from the noise reduction still gets conveyed pretty well.

The burden of proof is on the person making the initial positive claim. It's not everybody else's job to put in the effort to prove it isn't true. If you have a claim to make and you're trying to convince people of something you have to have evidence that's solid enough to convince skeptics.

Looks like once again with other people having to come in and do tests we see the same issues with cdng complained about and the softness or horribleness of braw not as bad as initially believed. Same thing different day, as expected.

As JB noted previously, the problem with this discussion began with strong negative words aimed at a codec many of us really like. I don't think any of us are arguing against allowing cdng as an option or allowing noise reduction settings if its possible. We are arguing that braw is not nearly as "bad"as the op suggests, that cdng isn't a savior or perfect and is often a hassle, and that his evidence was not convincing at all.

This is maybe the fourth or fifth time I've seen this play out with the same results.

If you're convinced cdng is perfect and BMD is a big meany for taking it away and is lazy or whatever for not doing what you want, you'll remain that way. And if you like braw for what it's designed for while also too agreeing improvement is always welcome when it's possible, you remain that way as well. Despite the evidence.

Same thing different day.

I would have dropped the VND as well because they are known to alter fine detail (even the good ones)
Again though the best way to test jacks claim is to replicate it with faces. The onus primarily being on him to do it because then it’s with the variables he has and uses.

JB

I've tested the B+W Vario that Robert and I use on higher res imaging setups than this (and with longer lenses which amplify the softness effect with vario ND's) and I couldn't see any. I've tested plenty of others that do soften the image hence why I only recommend these as vario ND's. I'd be confident in ruling this out as a factor. Ideally yes head shots would be best but then if we are digging down into the very limits of the sensors resolving power and looking at hairs it would be very difficult to replicate shots exactly for focus between setups esp if it involves firmware swops. I wonder if some of the tests that Jake or others have seen of 'live' subjects are similarly affected by very slight focus errors between setups?

The burden of proof is on the person making the initial positive claim. It's not everybody else's job to put in the effort to prove it isn't true. If you have a claim to make and you're trying to convince people of something you have to have evidence that's solid enough to convince skeptics.

Looks like once again with other people having to come in and do tests we see the same issues with cdng complained about and the softness or horribleness of braw not as bad as initially believed. Same thing different day, as expected.

As JB noted previously, the problem with this discussion began with strong negative words aimed at a codec many of us really like. I don't think any of us are arguing against allowing cdng as an option or allowing noise reduction settings if its possible. We are arguing that braw is not nearly as "bad"as the op suggests, that cdng isn't a savior or perfect and is often a hassle, and that his evidence was not convincing at all.

This is maybe the fourth or fifth time I've seen this play out with the same results.

If you're convinced cdng is perfect and BMD is a big meany for taking it away and is lazy or whatever for not doing what you want, you'll remain that way. And if you like braw for what it's designed for while also too agreeing improvement is always welcome when it's possible, you remain that way as well. Despite the evidence.

Same thing different day.

Well said.

If I remember last time (trigger warning) we went through this the testing methodology of the main claims were also in question.

And I still see those flawed tests being used as “evidence” of poor BRAW performance.

And as JG said, one persons “noise” is another’s “detail” once you get into subjective judgement.

JB

I would have dropped the VND as well because they are known to alter fine detail (even the good ones)
Again though the best way to test jacks claim is to replicate it with faces. The onus primarily being on him to do it because then it’s with the variables he has and uses.

JB

I've tested the B+W Vario that Robert and I use on higher res imaging setups than this (and with longer lenses which amplify the softness effect with vario ND's) and I couldn't see any. I've tested plenty of others that do soften the image hence why I only recommend these as vario ND's. I'd be confident in ruling this out as a factor. Ideally yes head shots would be best but then if we are digging down into the very limits of the sensors resolving power and looking at hairs it would be very difficult to replicate shots exactly for focus between setups esp if it involves firmware swops. I wonder if some of the tests that Jake or others have seen of 'live' subjects are similarly affected by very slight focus errors between setups?

All variable NDs are based on using polarisers.

Polarisers are all (like) very small diffraction grates. Like having to sets of picket fences. You rotate them to increase the amount of “blockage” or to let more light through.

But they are going to affect the resolution because it’s fundamental to how they work. And they affect it variably

Normally I’d say it’s insignificant. But if we’re really scrutinising around resolution it’s something that can easily be dropped. (Id use shutter speed for example or a regular known ND)

And yes, from memory one of the last times we went though this debate there was a shot that basically seemed to be out of focus (was it of Superman) I cant remember. But that’s why Jack you’re getting such a hard time around putting up evidence.

JB

Well, be nice, the guy's in a difficult position. OP came to forum under the impression he had big news to impart to BMD engineers ("I've achieved my mission of raising blackmagic's awareness").

But, having met opposition instead of acclaim, there's a problem: no convincing third-party proof of his claims, he won't accept well-executed tests which don't affirm his position (shoot rugs instead!) and he'll never be able provide test results himself, for obvious reasons -- ND or not, the claim is unsupportable, we've been through this a million times already.

Just another fine day in the forum....

The burden of proof is on the person making the initial positive claim. It's not everybody else's job to put in the effort to prove it isn't true. If you have a claim to make and you're trying to convince people of something you have to have evidence that's solid enough to convince skeptics.

Looks like once again with other people having to come in and do tests we see the same issues with cdng complained about and the softness or horribleness of braw not as bad as initially believed. Same thing different day, as expected.

As JB noted previously, the problem with this discussion began with strong negative words aimed at a codec many of us really like. I don't think any of us are arguing against allowing cdng as an option or allowing noise reduction settings if its possible. We are arguing that braw is not nearly as "bad"as the op suggests, that cdng isn't a savior or perfect and is often a hassle, and that his evidence was not convincing at all.

This is maybe the fourth or fifth time I've seen this play out with the same results.

If you're convinced cdng is perfect and BMD is a big meany for taking it away and is lazy or whatever for not doing what you want, you'll remain that way. And if you like braw for what it's designed for while also too agreeing improvement is always welcome when it's possible, you remain that way as well. Despite the evidence.

Same thing different day.

Man I really tried my best to not come across as one of those mindless braw haters, since I also read some previous forum posts about this, and their expectations were a bit too unrealistic. And they seemed to be crapping on braw as an entire codec. I‘m not crapping on braw, I’m just saying that the internal denoiser (used for both braw and prores) is a bit too extreme at times. I’ve also mentioned the benefits of braw like it’s speed, workflow, lack of compression artifacts etc. I just really used cdng is a benchmark for how much resolution the bmpcc4k can achieve without any processing. The amount of tests that I included were definitely limited and I do see how that affected my point. But I thought at least they were pretty obvious? I guess the subjective viewing experience also hindered that here. Cdng is nowhere near a perfect codec, but if braw can take the best of it and combine it with the advantages of it’s own. I think that would be great.

So here is the link to BRAW 3:1, ProRes and CDNG (one single frame still only - otherwise I would have had to reflash the firmware all the time without changing camera position).

https://we.tl/t-olJPILrESL

Well, be nice, the guy's in a difficult position. OP came to forum under the impression he had big news to impart to BMD engineers ("I've achieved my mission of raising blackmagic's awareness").

But, having met opposition instead of acclaim, there's a problem: no convincing third-party proof of his claims, he won't accept well-executed tests which don't affirm his position (shoot rugs instead!) and he'll never be able provide test results himself, for obvious reasons -- ND or not, the claim is unsupportable, we've been through this a million times already.

Just another fine day in the forum....

I really was just a lurker around this forum, and I’ve read some of the more extreme posts about this that demanded “the return of the cdng” and claimed braw as whole sucked, so I thought maybe asking for a user selectable level of denoising is more realistic and achievable? And I’ve said that Roberts tests were excellent, at proving braw and prores potential without the hinderance of the denoiser. And was the Superman the mini figurine test thing? I thought that test was a bit sketchy as well.

Huh, didn’t really notice that. Then I guess if denoising level could become user selectable that would be nice. Thank you for the info.

As couple of others have already said that it's not "noise reduction" as in an the application of an effect, it is the affect of the demosaic/compression algorithm. In this C5D review the BM guy specifically addresses the question of in camera noise reduction with Prores and Braw.

It's crazy, the moire and aliasing is painful but somehow we're looking for lost details in the shadow of a dumpster. I had to open the still in multiple SW to make sure it wasn't a scaling artifact. If anything the takeaway from your footage is the importance of a proper OLPF.

Good Luck

Huh, didn’t really notice that. Then I guess if denoising level could become user selectable that would be nice. Thank you for the info.

As couple of others have already said that it's not "noise reduction" as in an the application of an effect, it is the affect of the demosaic/compression algorithm. In this C5D review the BM guy specifically addresses the question of in camera noise reduction with Prores and Braw.

It's crazy, the moire and aliasing is painful but somehow we're looking for lost details in the shadow of a dumpster. I had to open the still in multiple SW to make sure it wasn't a scaling artifact. If anything the takeaway from your footage is the importance of a proper OLPF.

Good Luck

Wasn’t the quote specifically addressing how to get a better dynamic range readout by applying nr in resolve? And really I did initially thought that it was the demosaic and the compression as well, but why does it only smear the low contrast textures and underexposed areas? The well define textures and edges seemed to be fine, so that’s why I thought maybe it’s the denoising they apply before the demosaic in the camera.

As couple of others have already said that it's not "noise reduction" as in an the application of an effect, it is the affect of the demosaic/compression algorithm. In this C5D review the BM guy specifically addresses the question of in camera noise reduction with Prores and Braw.

This "email" appears to be talking about an improvement to the algorithm that BMD use to de-mosiac their raw footage. Its the same one that's in Resolve that's used on all RAW footage, and for a while back in the day, the camera's had a more simplified version of it so that you could monitor your image and for making ProRes. Then they were able to upgrade it to the same one that's in resolve and put that into the camera, around the same time I think that they "unified" the OS of the cameras.

Slightly different to the BRAW issues here, but an insight into how in-camera processing works. You know how you need all these GPU's to make a camera playback in real time ? Similar challenge in a camera. You need all the processing overhead to be able to do this in camera, and that was BMD acknowledging that they were now using the same algorithm. Previously in THEORY you could shoot CDNG and Process to ProRes in Resolve and maybe get a better result than shooting ProRes direct in camera.

Having the same algorithm in camera meant you were getting the SAME de-mosaic in camera as you would in resolve on a desktop.

All of this is pre-BRAW.

It's crazy, the moire and aliasing is painful but somehow we're looking for lost details in the shadow of a dumpster. I had to open the still in multiple SW to make sure it wasn't a scaling artifact. If anything the takeaway from your footage is the importance of a proper OLPF.
Good Luck

Well, yeah. One of artifacts of no OLPF are what we call "false" colour.

Look at the power line against the building in the OP's post. Look at how the colour fringes bothe red/ magenta and cyan / blue colours around it. That's not "real" information. It's an artifact of the sensor not being able to resolve that fine detail.

So we're trying to argue over if that kind of detail is important, despite the fact that it's "false".

It's not black and white, different degrees of detail trigger different degrees of artifact too.

So in BRAW, you have to draw a line under what you pre-filter and what you don't. Based on the data rate, the subjective image quality.

JB

Wasn’t the quote specifically addressing how to get a better dynamic range readout by applying nr in resolve? And really I did initially thought that it was the demosaic and the compression as well, but why does it only smear the low contrast textures and underexposed areas? The well define textures and edges seemed to be fine, so that’s why I thought maybe it’s the denoising they apply before the demosaic in the camera.

That was my bad, I misquoted. Dude was talking about dng/Prores, not Braw/Prores.

Good Luck

BRAW Q0 download link:
https://we.tl/t-DmH2nPcam9

Firstly, thanks Robert for re-doing the test that Jack should have done properly in the first place.

With your well documented methodologies if I was doing this I would have dropped the VND as well because they are known to alter fine detail (even the good ones)

I am currently locked down in New Orleans have don’t have any BMD cameras with me, only stills cameras or I would have also done this.

Again though the best way to test jacks claim is to replicate it with faces. The onus primarily being on him to do it because then it’s with the variables he has and uses.

Using other people’s footage is less useful. Jacks already I think said he would “pass” Roberts test with qualification. It’s bright and no faces ?

So again in circles.

I really think a lot of this would have ended with the right examples in the first place.

JB

I am also in a lockdown, with all of my previous tests being done in a production, so unfortunately I can not share it. And of course that's on me for not providing enough examples, but the jsfilmz sample basically demonstrates what I was talking about when I talked about muddiness on the far away faces and low contrast textures (https://drive.google.com/file/d/1HjxhVP ... pYS8M/view). And unfortunately the sigma 17-50 is all I got with me in this lockdown, but the low contrast smearieness from the noise reduction still gets conveyed pretty well.

John, if you actually read things instead of jumping to conclusions, which you do, you would have read Jack is locked down and his commercial footage is confidential, and you would know he specifically talked about the opposite situation that Robert shot. Other's footage tests are as scientifically relevant for whatever they are, except not what was asked about here. But I am happy for Robert's efforts. Why have you infuriatingly got to keep doing this? The right to talk is linked to talking rightly, but in this perverted over the top society, we can't even do that easily.

Don't expect big things unless you are disabled, be happy with the small threads. But in our perverse society it is the reverse. You all are shamefully putting this guy through the ringer and look how many thousands of users who are refusing to join you guys.

The burden of proof is on the person making the initial positive claim. It's not everybody else's job to put in the effort to prove it isn't true. If you have a claim to make and you're trying to convince people of something you have to have evidence that's solid enough to convince skeptics.

Looks like once again with other people having to come in and do tests we see the same issues with cdng complained about and the softness or horribleness of braw not as bad as initially believed. Same thing different day, as expected.

As JB noted previously, the problem with this discussion began with strong negative words aimed at a codec many of us really like. I don't think any of us are arguing against allowing cdng as an option or allowing noise reduction settings if its possible. We are arguing that braw is not nearly as "bad"as the op suggests, that cdng isn't a savior or perfect and is often a hassle, and that his evidence was not convincing at all.

This is maybe the fourth or fifth time I've seen this play out with the same results.

If you're convinced cdng is perfect and BMD is a big meany for taking it away and is lazy or whatever for not doing what you want, you'll remain that way. And if you like braw for what it's designed for while also too agreeing improvement is always welcome when it's possible, you remain that way as well. Despite the evidence.

Same thing different day.

Irrelevant and mistaken statements. You are not helping out your fellow man. He has seen proof you can add the opposite side, but nothing here has proved the contrary to the situation described, and the otherside grossly mistaken in many things (And Jack in certain things too). Read the following replies.

Irrelevant and mistaken statements. You are not helping out your fellow man.

Pot. Kettle. Black.

Maybe we should stick discussing the issue.

JB

Well, be nice, the guy's in a difficult position. OP came to forum under the impression he had big news to impart to BMD engineers ("I've achieved my mission of raising blackmagic's awareness").

Yes!

But, having met opposition instead of acclaim, there's a problem: no convincing third-party proof of his claims, he won't accept well-executed tests which don't affirm his position (shoot rugs instead!) and he'll never be able provide test results himself, for obvious reasons -- ND or not, the claim is unsupportable, we've been through this a million times already.

Just another fine day in the forum....

He did accept Robert's test, except the tests were of the wrong thing and therefore invalid. He should compare the shaded regions of those tests. I've asked if he could shoot walls or inside, or out the window. Shoot a neighbour firm the hall way standing in their door way maybe. So, the bright test doesn't really matter in this situation, but is completely relevant to bright performance. Pretty ob.

Repeated too many times, not by people like Jack.

The burden of proof is on the person making the initial positive claim. It's not everybody else's job to put in the effort to prove it isn't true. If you have a claim to make and you're trying to convince people of something you have to have evidence that's solid enough to convince skeptics.

Looks like once again with other people having to come in and do tests we see the same issues with cdng complained about and the softness or horribleness of braw not as bad as initially believed. Same thing different day, as expected.

As JB noted previously, the problem with this discussion began with strong negative words aimed at a codec many of us really like. I don't think any of us are arguing against allowing cdng as an option or allowing noise reduction settings if its possible. We are arguing that braw is not nearly as "bad"as the op suggests, that cdng isn't a savior or perfect and is often a hassle, and that his evidence was not convincing at all.

This is maybe the fourth or fifth time I've seen this play out with the same results.

If you're convinced cdng is perfect and BMD is a big meany for taking it away and is lazy or whatever for not doing what you want, you'll remain that way. And if you like braw for what it's designed for while also too agreeing improvement is always welcome when it's possible, you remain that way as well. Despite the evidence.

Same thing different day.

Man I really tried my best to not come across as one of those mindless braw haters, since I also read some previous forum posts about this, and their expectations were a bit too unrealistic. And they seemed to be crapping on braw as an entire codec. I‘m not crapping on braw, I’m just saying that the internal denoiser (used for both braw and prores) is a bit too extreme at times. I’ve also mentioned the benefits of braw like it’s speed, workflow, lack of compression artifacts etc. I just really used cdng is a benchmark for how much resolution the bmpcc4k can achieve without any processing. The amount of tests that I included were definitely limited and I do see how that affected my point. But I thought at least they were pretty obvious? I guess the subjective viewing experience also hindered that here. Cdng is nowhere near a perfect codec, but if braw can take the best of it and combine it with the advantages of it’s own. I think that would be great.

But for what it is, it is perfect, running at lossiess compression. It's Braw that isn't perfect, because it's designed to be lossy. Cdng sees exactly what the sensor is seeing at lossless, and Braw starts with this and hacks it. So, using lossiess Cdng start with what Braw reduces and you can do even more with it. We might all agree Braw is great for what it's good for, but a little extra can be done, which is ok to ask and not be grilled over a fire for it.

You are saying how bad Braw can get for your use case and it should have higher quality options and they say how bad Cdng is at lower quality setting, deliberately avoiding the comparable higher quality mode that I'm so dates their argument. It is an old trickster technique.

So here is the link to BRAW 3:1, ProRes and CDNG (one single frame still only - otherwise I would have had to reflash the firmware all the time without changing camera position).

https://we.tl/t-olJPILrESL

Would an external trigger do it? Like over Bluetooth?

Well, be nice, the guy's in a difficult position. OP came to forum under the impression he had big news to impart to BMD engineers ("I've achieved my mission of raising blackmagic's awareness").

But, having met opposition instead of acclaim, there's a problem: no convincing third-party proof of his claims, he won't accept well-executed tests which don't affirm his position (shoot rugs instead!) and he'll never be able provide test results himself, for obvious reasons -- ND or not, the claim is unsupportable, we've been through this a million times already.

Just another fine day in the forum....

I really was just a lurker around this forum, and I’ve read some of the more extreme posts about this that demanded “the return of the cdng” and claimed braw as whole sucked, so I thought maybe asking for a user selectable level of denoising is more realistic and achievable? And I’ve said that Roberts tests were excellent, at proving braw and prores potential without the hinderance of the denoiser. And was the Superman the mini figurine test thing? I thought that test was a bit sketchy as well.

Yep. It's not much use being reasonable, they take that like weakness and keep doing people over. The solution is to largely ignore them.

Huh, didn’t really notice that. Then I guess if denoising level could become user selectable that would be nice. Thank you for the info.

As couple of others have already said that it's not "noise reduction" as in an the application of an effect, it is the affect of the demosaic/compression algorithm. In this C5D review the BM guy specifically addresses the question of in camera noise reduction with Prores and Braw.

It's crazy, the moire and aliasing is painful but somehow we're looking for lost details in the shadow of a dumpster. I had to open the still in multiple SW to make sure it wasn't a scaling artifact. If anything the takeaway from your footage is the importance of a proper OLPF.

Good Luck

That embedded article quote hasn't turned up here. That sounds exactly like what I was saying in Dimitry's thread about extending dynamic range (which some didn't favour). Is there a link?

As I pointed out, Braw starts with exactly what cdng lossless produces, there is no gain as long as you can do the processing better outside Braw. Most of these arguments were never relevant, and what does the slowness of Cdng matter if your aim is to do what Braw doesnt, and you buy a more powerful computer system, and everybody else uses Braw in their existing systems like they did before. The anti vsc. I mean Cdng debate means even less when you realise the only thing asked is a little extra flexibility and better mode in braw alone.

So. Lossiess Cdng is about the veggie patch and cooking better. Braw is more about an end meal you change.

So, that quote is saying that it is done in the demosaicing etc. I never liked this sort of wide demosaicing years back, for reasons of what we are discussing here. It spreads light values. But also the aggressive compression smooths things out too. Both should be controllable to the level of effect on detail. Now the issue is it passed over a partial demosaicing that people presume is matching pixel values ofnyhe compression with underlying Bayer primary value at the underlying pixel location and smoothing out the other pixels in-between their primaries (which sounds a bit like my proposal for Elphel camera decade before last). Now we are told they scrap this and redo the debayering using the limited Bayer reconstructed values instead of expanding on the "partial" debayering. Notice the compression technique is equated with debayering, so does resolve expand on this technique or transfer to another format. So if compression technique is used as the basis of the finale debayer, that could be used as a spatial denoising technique producing the results here. So, does rendered out to a quality format still have the same issues, or better details, that can tell which it might be?

Wayne, Are you able to use your expertise to estimate what the data rates would be for a BRAW codec that retains more of the wholesome ingredients of CNDG and less of unpleasant artificial flavouring of BRAW?

Huh, didn’t really notice that. Then I guess if denoising level could become user selectable that would be nice. Thank you for the info.

As couple of others have already said that it's not "noise reduction" as in an the application of an effect, it is the affect of the demosaic/compression algorithm. In this C5D review the BM guy specifically addresses the question of in camera noise reduction with Prores and Braw.

It's crazy, the moire and aliasing is painful but somehow we're looking for lost details in the shadow of a dumpster. I had to open the still in multiple SW to make sure it wasn't a scaling artifact. If anything the takeaway from your footage is the importance of a proper OLPF.

Good Luck

Wasn’t the quote specifically addressing how to get a better dynamic range readout by applying nr in resolve? And really I did initially thought that it was the demosaic and the compression as well, but why does it only smear the low contrast textures and underexposed areas? The well define textures and edges seemed to be fine, so that’s why I thought maybe it’s the denoising they apply before the demosaic in the camera.

Ah, I keep forgetting to say. The way the compression works is it reduces the differences between surrounding pixeis to make it more compressible to the baseline. So this reduces surrounding contrast and muddies the picture. So, high contrast detail may suffer less, and low contrast detail more, plus it may target the low contrast as less noticeable.

@ Johns, no time to look at your replies. Just spent over 6 hours reading and writing and around 6 am. Such is life dedicated to some light. Come with me!

[/quote] In terms of dynamic range, colorscience, and artifacting braw is a great codec. But for our particular need of cropping it lacks a bit in the resolution department, that’s why I’m advocating for user selectable levels of denoising.[/quote]

This would have made a much less controversial albeit longer thread header ...and one that I completely agree is a reasonable request for some users

In terms of dynamic range, colorscience, and artifacting braw is a great codec. But for our particular need of cropping it lacks a bit in the resolution department, that’s why I’m advocating for user selectable levels of denoising.

This would have made a much less controversial albeit longer thread header ...and one that I completely agree is a reasonable request for some users

Yep definitely, I’m considering editing my original post to make it sound less “extreme”.

I will do some further tests with low contrast textures and in lowlight in the next days

I will do some further tests with low contrast textures and in lowlight in the next days

Sorry for making you do the work that I should've done mate, thank you very much for this. If you can also include some far away faces (like in the jsfilmz sample) with a fullbody shot of a person wearing same colored fabric as well that would be excellent, thank you so much.

Who thinks Jack and Wayne Steven are the same person? He gets to "make his points" twice as much this way. Conspiracy?

Who thinks Jack and Wayne Steven are the same person? He gets to "make his points" twice as much this way. Conspiracy?

I mean I guess we are on the same "scale" of opinions? I'd like to think braw is a decent codec, and that it's not braw's fault necessarily, it's the denoiser that gets applied to both braw and prores as well, so I'm really just targeting that. And plus Wayne seems to be around for quite a while now, and I'm really just a newbie lurker.

Wayne is a good, decent man. I enjoy reading his articles. His review is good. Every criticism about a product is good and (usually) contributes to improvement.

It's not necessarily just denoising at all. One of the principles of DCT compression is loss of detail where human vision doesn't expect it. We don't see small detail in low-contrast areas too.

Imagine a wall with fine grain painted white. You will simply see it as white when the light is soft. But if directional light is hitting at a flat angle, you'll see every little piece of grain from the same distance.

If that was the case I would've be happy, But it also smooths out fabrics, and sometimes faces as well, which is very jarring for the human eye to see. And plus prores is also compressed using dct, but when shooting prores on other cameras It always seems to be fine, only the prores and braw on the bmpcc4k and 6k has this aggressive smoothing problem. So I get that with dct you are going to lose something, but I feel like it's not nesscerary the codec's fault, since maybe the denoising before the demosaic can be too aggressive, which could carry 70% of the blame.