Comparison BRAW vs DNG (UMP G1)

Yesterday I setup in the studio some test scenes to make a direct comparison between BRAW and DNG.

Setup:
URSA Mini Pro Gen 1 / Sigma Cine 18-35mm T2.0 / Angelbird 256GB Cfast2.0 card
Firmware 6.0
Codecs: DNG lossless / BRAW Q0 / ProRes HQ / 4608 x 2592 px
Subjects lit by a Dedo DLED7-D with softbox, bounced into a Sunbouncer silver reflector.
For white balancing the X-Rite Colorchecker Video target was used.

Bluetooth+ App was used to switch between codecs and to start/stop recording to avoid any misalignment of the framing or focus between each recording.

I've imported then the files into Resolve 15.3.1 and just applied the Omeneo Primers EQX 04 Punch LUT, then created and exported a TIFF still of each shot. LUT is a technical LUT.
-> http://omeneo.com/primers/other/

While with BRAW there is definitely a slight loss of sharpness and details compared to DNG it can only be seen when enlarged to 200% or more.

One scene contained 10/20/50 Euro banknotes, which are a nice test subject because of having a lot of very fine print details. I have also shot a macro of a part of the 20 Euro banknote so you can see how the details really look like. I've tried to align the macro shot over the non-macro shots as good as possible.

These are 300% enlargements done in AFX with bicubic interpolation. The macro shot of course is not enlarged.

Euro_Banknotes_Detail_closer_BRAW

Euro_Banknotes_Detail_closer_BRAW.jpg (427.83 KiB) Viewed 23235 times

Euro_Banknotes_Detail_closer_DNG

Euro_Banknotes_Detail_closer_DNG.jpg (480.73 KiB) Viewed 23235 times

Euro_Banknotes_Detail_closer_Macro

Euro_Banknotes_Detail_closer_Macro.jpg (659.07 KiB) Viewed 23235 times

These are the full size shots - right click and select open image in a new tab for full res:

BRAW:


DNG

100% crops:

BRAW

Euro_Banknotes_Detail_BRAW.jpg (750.05 KiB) Viewed 23145 times

DNG

Euro_Banknotes_Detail_DNG.jpg (881.03 KiB) Viewed 23145 times

Downscaled Macro shot overlaid to the DNG version

Euro_Banknotes_Detail_Macro_overlaid.jpg (870.34 KiB) Viewed 23145 times

Shot of an old circuit board for Amiga - Cyberstorm 060:

Full size BRAW


Full size DNG


200% bicubic enlargements:

Cyberstorm_Details_BRAW

Cyberstorm_Details_BRAW.jpg (523.68 KiB) Viewed 23139 times

Cyberstorm_Details_DNG

Cyberstorm_Details_DNG.jpg (619.44 KiB) Viewed 23139 times

You can see the false details and artefacts of the DNG debayering in the diagonal lines.

Shot of a pair of working gloves:

Full size BRAW


Full size DNG


200% bicubic enlargements:

Gloves_Detail_BRAW.jpg (422.43 KiB) Viewed 23134 times

Gloves_Detail_DNG.jpg (493.53 KiB) Viewed 23134 times

Interestingly that with BRAW the gray texture had a slight drop in the blue channel, so it looks a bit more reddish than the DNG.

Shot of Hagomoro Luminous Color chalks:

Full size BRAW


Full size DNG


200% bicubic enlargements:

Hagoromo_Chalks_Detail_BRAW.jpg (284.32 KiB) Viewed 23126 times

Hagoromo_Chalks_Detail_DNG.jpg (387.7 KiB) Viewed 23126 times

BRAW did change the texture of the blue channel quite significantly.

Interesting to also note the "jaggies" or "zippers" on diagonals in the closeups from cDNG. Look at the macro still and braw... no apparent jaggies. This is something I had to deal with consistently when shooting cDNG. I could see those issues in full view and would blur the image to remove them.

A little off topic...

Have you noticed a benefit of shoot 3:1 instead of Q:0 on shots with no movement? I have noticed that 3:1 on the pocket 4K gives a slight bit more detail compared to Q:0 in these situations, and Q:0 providing a little more detail when shooting something like a wide shot with lots of movement.

Strange how it influences the texture on the chalk tip:

Hagoromo_Chalks_Detail_BRAW_blue_channel.jpg (291.64 KiB) Viewed 23119 times

Hagoromo_Chalks_Detail_DNG_blue_channel.jpg (417.56 KiB) Viewed 23119 times

Shot of some miniatures:

Full size BRAW


Full size DNG


200% bicubic enlargements of detail 1:

Minatures_Detail1_BRAW.jpg (379.92 KiB) Viewed 23107 times

Minatures_Detail1_DNG.jpg (460.1 KiB) Viewed 23107 times

Note the strange jaggies in BRAW around the red areas of the sword.

200% bicubic enlargements of detail 2:

Minatures_Detail2_BRAW.jpg (320.67 KiB) Viewed 23107 times

200% bicubic enlargements of detail 2:

Minatures_Detail2_DNG.jpg (412.94 KiB) Viewed 23107 times

200% bicubic enlargements of detail 3:

Minatures_Detail3_BRAW.jpg (400.9 KiB) Viewed 23107 times

Minatures_Detail3_DNG.jpg (482.08 KiB) Viewed 23107 times

Red channel of BRAW has some strange halo and jaggies:

Minatures_Detail1_BRAW_red_channel.jpg (351.61 KiB) Viewed 23098 times

Minatures_Detail1_DNG_red_channel.jpg (408.97 KiB) Viewed 23098 times

I'd say there is definitely some room for improvement in BRAW processing behavior.

Thanks, Robert, for your illuminating attention to the details. So there are some noticeable differences in read and blue when shooting with the Mini. Since BRAW is calibrated for each sensor, would be very interesting for someone with a BMPCC4K that can illustrate BRAW video and perhaps a DNG still to see if these differences in edge texture and colour are evident in a controlled studio setting.

I think the bottom line is, unless you're a pedant, no one will really see any tangible difference between BRAW and CDNG in the real world, that is, looking at a video. I am assuming that people produce videos for viewing as a video and not to produce still image to be enlarged over 100%.

DSMC(2) (Digital Still and Motion Camera) on a budget.

Ray, I would agree that for normal motion work BRAW is great.
But what me concerns is the behavior in the reds and some of the strange halos of my samples which would pose a problem with greenscreen work. This has to ironed out by BMD and I have to admit to what others here have written in the past - for some VFX work you would still be better of with DNG. At least you should do some initial testing for a particular job, otherwise you might be have to be prepared for surprises.

Thanks Robert,

Lots of work went into this I'm sure, thanks for posting.

I'm between continents right now, but I wonder if it makes a difference to have this test IN MOTION.

We never look at 200% blow ups of still frames, but I wonder if they were 5 sec loops in motion how our perceptions would change.

JB

Robert is onto something that I’m sure BMD will be investigating. That pair of frames that shows an apparent drop in the blue channel almost looks like infrared contamination, but it’s not. If his setup is still in place, perhaps Robert can include a couple of clips with some motion such as using his slider or panning.


Sent from my iPhone using Tapatalk

I’ve definitely found a drop in the blue channel (less reddish) on the V4 CS on UMP

I think the bottom line is, unless you're a pedant, no one will really see any tangible difference between BRAW and CDNG in the real world, that is, looking at a video. I am assuming that people produce videos for viewing as a video and not to produce still image to be enlarged over 100%.

Amen!

And yeah, great job Robert.

Excellent comparisons, Robert! Many thanks!

Thanks Robert. Good test. BRAW looks a bit softer. This may be a limit of the new codec for film shooting. For big cinema screen CDNG looks still a better option. BM should in future improve this difference, in case if CDNG have been already included as a "white list" of new cameras (BMPCC4K).

For big cinema screen CDNG looks still a better option.

Even assuming more detail is better, or of any dramatic consequence, these differences are far too small to discern in a typical commercial movie theater. It's only at home, on still frames, a few millimeters from the screen, that this sort of thing gets people excited.

I'm between continents right now, but I wonder if it makes a difference to have this test IN MOTION.

We never look at 200% blow ups of still frames, but I wonder if they were 5 sec loops in motion how our perceptions would change.

I was thinking about how to accomplish a comparable test in motion. As I do not have a motion control rig, it would be impossible to move the camera in a repeatable way.

But what I have is a turntable. So I have setup today a new scene with the 3 miniatures on the turntable and recorded a full loop with each codec (BRAW Q0 / BRAW C 3:1 / CDNG lossless).

In Resolve I tried to line up the three shots as close as possible and rendered out the same 11 sec segment of each version to 4608 x 2592px Cineform YUV 10-bit 25fps.
I also tried to render out to Cineform RGB 12-bit - but I could not see any differences, so I went with the smaller file sizes and easier to play YUV encoding.

Because I cannot host 4 GB files, I am uploading them to WeTransfer where they will be online for a week. I can re-upload again later anytime if people want.

So here are the 3 clips:

4608 x 2592px Cineform YUV 10 bit BRAW Q0 and C 3:1
https://we.tl/t-KGiDHN2fdI

4608 x 2592px Cineform YUV 10 bit DNG lossless:
https://we.tl/t-7nBz9vz2qt

Online til May, 30th 2019 / 16:00 CEST

Remember: You will need an installed Cineform codec to play the clips.

EDIT: active link without an expire date:
https://we.tl/t-bmyDwUzV9f

Thanks for share. really great tests!

So here are the 3 clips:

4608 x 2592px Cineform YUV 10 bit BRAW Q0 and C 3:1
https://we.tl/t-KGiDHN2fdI

4608 x 2592px Cineform YUV 10 bit DNG lossless:
https://we.tl/t-7nBz9vz2qt

Online til May, 30th 2019 / 16:00 CEST

Remember: You will need an installed Cineform codec to play the clips.

Care to share your observations ? Does motion make any difference to the sharpness perception?

JB

I have put all three clips into the playlist of MPC-HC and started to let them play in an endless loop. Waited for a minute without watching so I didn't know which clip was which.

My observations:

Watching the clips downscaled to 1080p on a 24" 1920 x 1200 monitor - no differences can be seen whatsoever. No surprise here.

Watching the clips downscaled to fully height fit into a 34" 3440 x 1440 monitor - no differences can be seen whatsoever.

Watching the clips downscaled to fit into the width of a 34" 3440 x 1440 monitor - tiny differences can be seen - but only if you know where to look for and if you have the possibility to compare to the DNG version. If the clip would be BRAW only - nothing to complain.

Watching the clips at actual 1:1 pixel size on the 34" 3440 x 1440 monitor - differences can be seen if you are used to see 4k. I had no problem to single out the DNG clip anytime out of the three. Again, I have a trained pixel peeper eye, was watching a loop, and knew where to look for differences.

I would say if you record full sensor on the UMP and deliver in UHD, you will never see the difference because you started with an oversampled image.

On the PCC4k this would be another story because you can only record a 1:1 image.

Worst case will be windowed HD on any of the cameras.

When last year I tested my PCC4k in 120fps mode (ProRes) - I was quite surprised that it looked softer than the 120fps from my UM46k. I didn't know back then, that it already got the new debayering, but I immediately saw the difference.

I think BMD can and has to do some improvements on the noise reduction and the way it detects edges. The color haloing on primary color borders is something that needs to be addressed, especially for bluescreen work. It reminds me of the color bleeding of the old 4:1:1 DV codecs. It might be less problematic with greenscreen - but I haven't done any tests on that yet.

This thread is so informational. Much thanks to everyone contributing.

PS. It would be cool if BM would at least confirm that 'hey, we know it's an issue and we're looking into it.'

To see better how and which colors are affected by halo in braw, you can shoot color checker and look around every color patch. This problem was visible in braw in every first test samples done on 4.6K camera. So yes, BM really need to find a way how to fix it.

The same halos are seen with Prores, it's not unique to braw, for whatever that means.

As i remember those early test images where only BRAW and DNG. Compare color halo with ProRes also may be useful.

I don’t foresee using the BMPCC4K HD at 120 fps. Therefore am I correct to takeaway from this thread that I’ll get the best HD deliverables by recording in UHD BRAW in camera and downscaling in post to HD?

I’m hoping to avoid recording in ProRes HD in camera from an UHD sensor scan. But would that be preferable to recording ProRes HD in camera windowed?


Sent from my iPhone using Tapatalk

Yes, I would think so Rick, as the UHD window down sampled to HD will have more detail (even in the in camera down converted recording), and less noise. First choice would be recording in UHD, down sample in post, followed by recording in camera HD from UHD window, third would be the HD window (which will have more noise, fewer pixels getting the info).
Cheers

On an HD monitor, from an HD timeline, I can't tell the difference between BMPCC 4K Prores down-scaled in camera (from full sensor to HD) and UHD braw downscaled to HD in Resolve. There *is* a difference -- zoomed in, the UHD braw is sharper than the full sensor camera downscale -- but of course it's also twice the resolution, though this is far less obvious than you might think. The Prores footage seems to have less noise than braw, so that's a complicating factor.

Windowed HD is softer than either down-scaled version, and noisier.

For big cinema screen CDNG looks still a better option.

Even assuming more detail is better, or of any dramatic consequence, these differences are far too small to discern in a typical commercial movie theater. It's only at home, on still frames, a few millimeters from the screen, that this sort of thing gets people excited.

I think that to correct this critical resolution and micro contrast of new codec BRAW (close to CDNG) is important at least from marketing point of view. I mean to open this option to end users. (If BM put CDNG in white list.) This will close all questions.

As i remember those early test images where only BRAW and DNG. Compare color halo with ProRes also may be useful.

The color halo looks identical on my ProRes shots. I have done the tests with all flavors of the codecs, but didn't post the ProRes results because they were too similar to the BRAW results.

The color halo looks identical on my ProRes shots. I have done the tests with all flavors of the codecs, but didn't post the ProRes results because they were too similar to the BRAW results.

Which seems to be a compression artefact ? ProRes is DCT based, and I think this is the same thing here..an artefact of compression.

There was a thread here a while ago where someone’s discovered a similar CDNG compression artefact as well on very fine detail, like tree branches against a high contrast sky BG.

It’s important to remember that this is going to happen with any codec when you use compression. You don’t throw away a bunch of the image data without some kind of cost somewhere.

JB

The color halo looks identical on my ProRes shots. I have done the tests with all flavors of the codecs, but didn't post the ProRes results because they were too similar to the BRAW results.

Which seems to be a compression artefact ? ProRes is DCT based, and I think this is the same thing here..an artefact of compression.

John, that is 100% not a compression artifact. I am dealing with all kind of codecs and compression algorithm since 2000 and I know exactly what's an artifact of DCT compression and what is not.
Back in the early days I had dived deep into MPEG, MPEG2, Divx and XviD compression schemes and the countless parameters you could tweak for getting optimized encoding quality...

Ringing, mosquito noise and block artifacts would be a result of DCT.
What we can see here is definitely a problem of the noise pre-processing done in the camera. And maybe also of the debayer algorithm. Some of the parameters of the spatial denoiser in the camera need to be tweaked. You can see what's going on clearly when inspecting the single color channels.

which would pose a problem with greenscreen work.... for some VFX work you would still be better of with DNG.

who keeps saying this because this is not true. The most important part to key a greenscreen is if it is lit correctly. I keyed greenscreen from red footage and from the c100 MK2. C100 MK2 I was the vfx supervisor for. So i personally made sure there were minimum green spill. However, for the red footage it was greenspill everywhere. It had a easier time keying the c100 than the Red. Braw should handle any vfx you throw at it, the biggest difference between CDNG and Braw is its sharpness and that's not going to mess up any vfx work.

Thanks Robert. Good test. BRAW looks a bit softer. This may be a limit of the new codec for film shooting. For big cinema screen CDNG looks still a better option. BM should in future improve this difference, in case if CDNG have been already included as a "white list" of new cameras (BMPCC4K).

I could tell you the opposite
Big screen need less sharp picture, first you cannot see phisically (see the study of schnellen about resolution perception of human eyes with 10/10) a real sharpness of 4K on big screen, to see details of white point and black point on 30 meters of screen you need to be at 2.75 meters near the screen, but you cannot see entire screen.
Big cinema screen had low contrast due the reflection of light, then you see less details.
But
If you see too much sharp picture at 24 FPS on 30 meters screen often you will see the strobo on motion, if you put same picture on 50 inc uhd tv you not have same problem of strobo, every frame is changed on small scale.

More sharpness is a mental sickness of recent era, sold from marketing of tv, ultra definition and more, but not ever is good especially for movie at theater for a good experience.

Next time you go to cinema try to go near the screen during commercial, during trailer and try to see that sharpness...

Why movie camera are so different from video camera? They not have sharpness addition (halo and edging) to false the details like video camera
More soft be cause movie is soft experience

Ps if someone tell me that Ursa Mini Pro and pocket have sharpness controllo I answer that also in Blackmagic Design manual is stated that is artificial addition ONLY for live use of that cameras.

Dng could have more details? Yes
Does it matter? Yes and no
If you need to extract a photo from a frame could be better, you must shoot for a photo, with different shutter angle and different composition then is a different use of camera.
If you need to shoot motion picture shoot in Q0 and in 3:1 and you all quality that you need.
In big screen goes bad movie shooted in h264 intra and in the other worst codec for color correction, but go on screen. Amen.

Anyway in past many movie we’re shooted in 16mm and pumped to 35mm.
Today we can shoot at4-4.6k to delivery a excellent 4K-2k and we are here to discuss of microdetails at freeze frame at 300%...

I go back to write my screenplay, if I will write bad also if I shoot it in 16k stereo 3d in odorama it will be a bad products



Inviato dal mio iPhone utilizzando Tapatalk

I go back to write my screenplay, if I will write bad also if I shoot it in 16k stereo 3d in odorama it will be a bad products

The joke here is that high resolution on low and ultra-low budget productions absolutely murders the movie. It's impossible to sustain dramatic illusion when the faults of cheap hair, makeup, set design and lighting are seen in all the glory of 4K (and HD is already bad enough).

I go back to write my screenplay, if I will write bad also if I shoot it in 16k stereo 3d in odorama it will be a bad products

The joke here is that high resolution on low and ultra-low budget productions absolutely murders the movie. It's impossible to sustain dramatic illusion when the faults of cheap hair, makeup, set design and lighting are seen in all the glory of 4K (and HD is already bad enough).

that's right, i remember years ago friend of mine that he is scenographer that hate the coming of HD in tv, all work to redo, everythings that work fine in sd in hd show defects, from a simple paint stroke on background to bad makeup on tv presenter ...

More sharpness is a mental sickness of recent era, sold from marketing of tv, ultra definition and more, but not ever is good especially for movie at theater for a good experience.

Next time you go to cinema try to go near the screen during commercial, during trailer and try to see that sharpness...

Why movie camera are so different from video camera? They not have sharpness addition (halo and edging) to false the details like video camera
More soft be cause movie is soft experience

I do not speak about sharpness itself. I say only that technically more information are available out of the sensor your codec able to operate - better and more flexible post production. How footage will look like on screen it's a deal of the end user.

I think that main difference of movie and video camera is not sharpness itself but first (I think) dynamic range.

which would pose a problem with greenscreen work.... for some VFX work you would still be better of with DNG.

who keeps saying this because this is not true. The most important part to key a greenscreen is if it is lit correctly. I keyed greenscreen from red footage and from the c100 MK2. C100 MK2 I was the vfx supervisor for. So i personally made sure there were minimum green spill. However, for the red footage it was greenspill everywhere. It had a easier time keying the c100 than the Red. Braw should handle any vfx you throw at it, the biggest difference between CDNG and Braw is its sharpness and that's not going to mess up any vfx work.

Chris, I have done VFX work myself, especially greenscreen. I have pulled keys from 4:2:0 MiniDV and HDV, from C100, FS7, 16mm film, BMCC, and UM46K. But this is not about a badly lit greenscreen vs correctly lit greenscreen. Just download the 11 sec sample footage I have provided here and look at the red part of the sword in the BRAW versions. You can easily see the problem there versus the DNG version providing a clean border between the red part and blue background.

I do not speak about sharpness itself. I say only that technically more information are available out of the sensor your codec able to operate - better and more flexible post production. How footage will look like on screen it's a deal of the end user.

I think that main difference of movie and video camera is not sharpness itself but first (I think) dynamic range.

I agree with you, dynamic range do the difference, if you can see the alexa sensor, stay on 2.7 (or 3.5 in opengate) for many years, but good enought to shoot a lots of motion picture.
The main difference from Alexa to bmd is not a global dynamic range, but the real dectail in all step of dinamic range.
On ursa mini pro you can use since 8-9 step of DR, if go down you reach FPN, if you go up you have some color shifting.
On Arri you can push all step without so big problem, less resolution more clean picture.
Then... like an old Dp teach me, film strip had less dr than actual digital sensor and they did movie for years, all that claiming about the needings of DR often is marketing.
And he was the Dop of Inferno and Phenomena's Dario Argento, two film with a very strange and complex filmography.
He teach me to learn the limits of my tools, use them and go around it.
We have great tools in our hands, also with actual DR

I'm a film man. Shooting film for many decades but in still photography, printing bw silver gelatin. I use point light and know how it works in final print (let say vs diffused light source) (Difference is very close to CDNG vs BRAW). I understand in very deep details what micro contrast means and how it works in graphics. Even if you think that your eyes do not see (or not pay attention) to any fine details/film grain your brain (all the times) evaluate this matter. Your brain understand are there any more information inside of the frame or not. My approach more information to be better in any case (even if do not see it at the moment it works).

ps I prefer good bw film copy (old classic films) and first row in cinema theater. And if let say one machine is in perfect critical grain focus and another is a bit out (only grain of film copy). This difference between parts of the film is visible. But in this case you are able to compare this matter several times within at least 90min. Normally I'm not happy out of this matter.))

Thank you for posting this Robert. Pictures tells thousand words, and I agree in much of what you are saying. Also found much the same by my own tests.

The whole community can discuss this to the worlds end, it seems. BRAW will get better in time though, I strongly believe that. It was the same steps with red code, a lot of commotion in the beginning.

Anywho... after working with BRAW in many various conditions. I get this question in my head.

Who, in their right mind, will raise their hand telling that they can’t produce professional and beautiful imagery with BRAW? Today...in it’s current state. I’m just curious...

Thank you very much for this extreeeeeeeeeeme amount of work you have put into it. I think we all owe you big time for this.

In the codecs you wrote: DNG lossless / BRAW Q0 / ProRes HQ

But I dont see the ProRes HQ variant in this comparion?

We can say this and that but at the end, the audience don.t give a sht, The story is the only point.


Ricardo Marty

Thank you very much for this extreeeeeeeeeeme amount of work you have put into it. I think we all owe you big time for this.

In the codecs you wrote: DNG lossless / BRAW Q0 / ProRes HQ

But I dont see the ProRes HQ variant in this comparion?

I didn't include ProRes HQ screenshot because they were indistinguishable to the BRAW samples and wouldn't have added anything to the comparison.

Several months after this thread began I wonder if positions have coalesced on the best current practices. Would you kindly comment on which codecs and resolutions you feel would be best for my upcoming music video?

Final deliverables will be either 2K or HD. Want to be able to use both cameras, one shooting BTS and one the music video.

URSA Mini 4.6K - the Music Video:
maximum capture UHD within 1.33x Anamorphot Adapter for 2.39:1 Timeline;
interior and exterior Day;
- should I use CinemaDNG 3:1 24 fps or ProRes 444?

BMPCC4K - the BTS:
could be 4K DCI or UHD, cropped to Widescreen 2.39:1;
interior but might have exterior Day;
- should I use BRAW or ProRes 422 HQ?

Resolve 16 timeline using maximum resolution to Deliver 2K/HD.

I keep vacillating what will work the best. No euros will be harmed or used in this video!

I’d like to use the raw on each camera, but not if that’s a recipe for disaster. Thank you very much for your input.




Sent from my iPhone using Tapatalk