Topics

BRAW tests

Geoff Boyle
 

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

Jeff Kreines
 

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

Sent from iPhone. 

On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

Colin Elves
 

CDNG is the most obvious choice since that’s the other Raw format the UMP can encode directly. 

I guess the obvious choices are things that might cover: resolution, noise, colour resolution - and anything that might throw up debayering errors. Some people have suggested BRAW is a YCRCB codec as there is some ghosting in the red channel at high contrast edges.

Colin Elves
Director of Photography
UK Mob: +44 7886 849 073
German Mob: +49 157 78356204




On 2 Oct 2018, at 08:56, Jeff Kreines <jeff@...> wrote:

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

Sent from iPhone. 

On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

Geoff Boyle
 

CDNG is a very good idea, at all the variations that are available in camera.

 

Last time I looked Cineform wasn’t an included option, perhaps I missed something in my haste to check out the new facilities in the latest firmware.

 

Now back to my original question…

 

What subject will show the effects of compression the most?

 

Bear in mind that it has to be simple, obvious and easily available.

 

I’ve been looking at the trees outside my window blowing in the wind. Lots of motion, fine detail, huge mix of light and dark, can be shot so that the entire image is changing every frame.

 

cheers
Geoff Boyle NSC
EU based cinematographer
+31 637155076

www.gboyle.nl

www.cinematography.net

 

 

From: cml-raw-log-hdr@... <cml-raw-log-hdr@...> On Behalf Of Colin Elves
Sent: 02 October 2018 10:08
To: cml-raw-log-hdr@...
Subject: Re: [cml-raw-log-hdr] BRAW tests

 

CDNG is the most obvious choice since that’s the other Raw format the UMP can encode directly. 

 

I guess the obvious choices are things that might cover: resolution, noise, colour resolution - and anything that might throw up debayering errors. Some people have suggested BRAW is a YCRCB codec as there is some ghosting in the red channel at high contrast edges.

 

Colin Elves

Director of Photography

UK Mob: +44 7886 849 073

German Mob: +49 157 78356204



 

 


On 2 Oct 2018, at 08:56, Jeff Kreines <jeff@...> wrote:

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

 

Sent from iPhone. 


On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

Mark Kenfield
 

Leaves swaying in the breeze and gurgling water have always been my go-tos for testing compression.

Cheers,

Mark Kenfield
Cinematographer

0400 044 500


On Tue, 2 Oct 2018 at 22:56, Geoff Boyle <geoff.cml@...> wrote:

CDNG is a very good idea, at all the variations that are available in camera.

 

Last time I looked Cineform wasn’t an included option, perhaps I missed something in my haste to check out the new facilities in the latest firmware.

 

Now back to my original question…

 

What subject will show the effects of compression the most?

 

Bear in mind that it has to be simple, obvious and easily available.

 

I’ve been looking at the trees outside my window blowing in the wind. Lots of motion, fine detail, huge mix of light and dark, can be shot so that the entire image is changing every frame.

 

cheers
Geoff Boyle NSC
EU based cinematographer
+31 637155076

www.gboyle.nl

www.cinematography.net

 

 

From: cml-raw-log-hdr@... <cml-raw-log-hdr@...> On Behalf Of Colin Elves
Sent: 02 October 2018 10:08
To: cml-raw-log-hdr@...
Subject: Re: [cml-raw-log-hdr] BRAW tests

 

CDNG is the most obvious choice since that’s the other Raw format the UMP can encode directly. 

 

I guess the obvious choices are things that might cover: resolution, noise, colour resolution - and anything that might throw up debayering errors. Some people have suggested BRAW is a YCRCB codec as there is some ghosting in the red channel at high contrast edges.

 

Colin Elves

Director of Photography

UK Mob: +44 7886 849 073

German Mob: +49 157 78356204



 

 


On 2 Oct 2018, at 08:56, Jeff Kreines <jeff@...> wrote:

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

 

Sent from iPhone. 


On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

Nick Shaw
 

On 2 Oct 2018, at 14:51, Mark Kenfield <mark@...> wrote:

Leaves swaying in the breeze and gurgling water have always been my go-tos for testing compression.

If the compression is inter-frame like h.264, sure. But I had assumed (although I don't know if this is confirmed) that BRAW is an intra-frame compression scheme, in which case movement is not relevant, except insofar as it may add motion blur to the images, which actually helps compression.

Nick Shaw
Workflow Consultant
Antler Post
UK

+44 (0)7778 217 555

Paul Curtis
 



On 2 Oct 2018, at 15:20, Nick Shaw <nick@...> wrote:
On 2 Oct 2018, at 14:51, Mark Kenfield <mark@...> wrote:
Leaves swaying in the breeze and gurgling water have always been my go-tos for testing compression.
If the compression is inter-frame like h.264, sure. But I had assumed (although I don't know if this is confirmed) that BRAW is an intra-frame compression scheme, in which case 

I would still say Leaves are better than water. High shutterspeed and you'd be able to easily see the compression artefacts inside a frame. There are some great examples with R3D showing the way increasing compression changes the image (using distant leaves).

If you shoot with uncompressed DNG (Is that possible) and then some compressed variants of BRAW then that would be a great control/baseline for checking compression but also debayer sharpness. You can take a single DNG as a master and process that to see what kind of flexibility you loose with BRAW, if any.

The rest is colour stuff and that's usually eyeballing and feeling. You could shoot way off on white balance to see whether you can fix that afterwards like RAW. The question is what are you missing with BRAW.

There was an interesting interview in the latest F&D times where Jarred @ Red implied they licensed some aspects of compressed RAW to Apple. Wasn't *directly* said but that was the implication. I presume BRAW hasn't licensed anything hence this approach.


cheers
Paul

Paul Curtis
VFX & Post | Canterbury, UK

Luis Gomes
 

Taking a big risk by sharing my opinions here on this distinguished forum. 
Shot my girlfriend (black long hair) against a very bluish (Finnish sunset) sky. 
Absolute horrible artifacts of all sort. 

Also. All my tv host with their great very thin pinstripes suits. Never fails me. 

Luís 
Finland 
Camera but nothing fancy :-)
--
Gomes.luis@...
http://fi.linkedin.com/pub/luis-gomes/20/11b/335/
Freelancer video Professional. 
Finland. 

Robert
 

You could shoot a row of kids pinwheels spinning in front of a chain link fence with trees behind. Do a short slider move or track r or l - almost guaranteed to break the codec. 

Robert Goodman
Cinematographer/Writer
newly retired Professor 
Copenhagen Denmark
+45 21 85 63 13



On Oct 2, 2018, at 3:51 PM, Mark Kenfield <mark@...> wrote:

Leaves swaying in the breeze and gurgling water have always been my go-tos for testing compression.

Cheers,

Mark Kenfield
Cinematographer

0400 044 500


On Tue, 2 Oct 2018 at 22:56, Geoff Boyle <geoff.cml@...> wrote:

CDNG is a very good idea, at all the variations that are available in camera.

 

Last time I looked Cineform wasn’t an included option, perhaps I missed something in my haste to check out the new facilities in the latest firmware.

 

Now back to my original question…

 

What subject will show the effects of compression the most?

 

Bear in mind that it has to be simple, obvious and easily available.

 

I’ve been looking at the trees outside my window blowing in the wind. Lots of motion, fine detail, huge mix of light and dark, can be shot so that the entire image is changing every frame.

 

cheers
Geoff Boyle NSC
EU based cinematographer
+31 637155076

www.gboyle.nl

www.cinematography.net

 

 

From: cml-raw-log-hdr@... <cml-raw-log-hdr@...> On Behalf Of Colin Elves
Sent: 02 October 2018 10:08
To: cml-raw-log-hdr@...
Subject: Re: [cml-raw-log-hdr] BRAW tests

 

CDNG is the most obvious choice since that’s the other Raw format the UMP can encode directly. 

 

I guess the obvious choices are things that might cover: resolution, noise, colour resolution - and anything that might throw up debayering errors. Some people have suggested BRAW is a YCRCB codec as there is some ghosting in the red channel at high contrast edges.

 

Colin Elves

Director of Photography

UK Mob: +44 7886 849 073

German Mob: +49 157 78356204



 

 


On 2 Oct 2018, at 08:56, Jeff Kreines <jeff@...> wrote:

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

 

Sent from iPhone. 


On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

axel.mertes
 

Hi Geoff,

first of all, BRAW is surely an i-frame only codec, which means it should be absolutely irrelevant if you have a fast moving scene or not when it comes to the result of each single frames compression. Each frame is compressed individually without any relation to previous or following frames. That makes it fast on decoding, as you can jump anywhere in the timeline and do not need to wait for decoding previous or later frames to decode the actual image (which is the case with any GOP based codecs like MPEG1/2/4, H265 etc.). As Nick & Paul already pointed out, motion blur will actually help compression, as its a natural low frequency smoothing smear over otherwise high frequency detail. So it might be interesting to use a typical 180° shutter and - only additional if you find it useful - forcefully a very short shutter for ultra-sharp images (we know you don't like them anyway).

To me the typical 180° shutter is way more interesting than the hypothetical ultra sharp short shutter, because if the tests tells us later that THIS is going to be a bad combination with compression, we just learn that we should shoot this in CDNG or uncompressed then...

What might show off codec compression artefacts is usually high noise/frequency content, i.e. grass, leaves, gravel, windy water, moire-effect forcing "check patterns" or "pin stripe suit" silks. Then of course green sceen or blue screen with fine contour details like single hair, wavy curtains with embroidery, or fine grass in front of it. Even a not to close close up to really gray hair (should be available...) on a "hair thickness is not much more than 1-3 photosites" is often good to reveal color interpolation (demosaic) issues. Thats the stuff where we see it in post. The finer the detail, the more likely its hard to reproduce with compression. Each compression scheme tries to reduce high frequency noise - thats where the details go...

CDNG is very interesting in that context, as is ProRes444 highest settings (at least).

A stable calm set might seem better than a highly changing set, as you can not 100% compare the frames. But I agree, we are interested in motion artefacts visible due to compression. As the compression is frame by frame, it might hit each frame a bit differently and might show off artefacts that are less visible in each the stills than in motion sequence. As its obviously a DCT codec, we should have an eye on if we can see/reveal grids at some point (aka JPEG artefact), preferably using an excessive gain in post.

Dynamic compression ratio (quality based, so more complex frames will be bigger) can be expected to show off a constant quality of artefacts (if any).
Fixed data rate settings in turn will show way more artefacts on fine detail content than on smooth and easy content.

What I also would like to see - if you can manage it - is a clear dark blue sky horizon. Something where the gradient in blue is very very small and subtle. I have a clip from a RED camera (not shot by myself) that is perfect for showing what the difference of an 8 bit vs. a 10 bit display is like. You see color banding on 8 bit displays, while its super smooth on 10 bit displays. That is, because there is very low noise in it. If I dither down to 8 bit (distributing the error by "intelligently" adding the error as kind of noise) you can make it look nearly as good on 8 bit as on 10 bit - but you file is less good compressable. Here is were you learn that low end distribution codecs have their own set of problems...

Mit freundlichen Grüßen,
Best regards,

Axel Mertes

Workflow, IT, Research and Development
Geschäftsführer/CTO/Founder
Tel: +49 69 978837-20
eMail: Axel.Mertes@...

Magna Mana Production
Bildbearbeitung GmbH
Jakob-Latscha-Straße 3
60314 Frankfurt am Main
Germany
Tel: +49 69 978837-0
Fax: +49 69 978837-34
eMail: Info@...
Web: http://www.MagnaMana.com/

Diese E-Mail enthält vertrauliche und/oder rechtlich geschützte Informationen. Wenn Sie nicht der richtige Adressat sind oder diese
E-Mail irrtümlich erhalten haben, informieren Sie bitte sofort den Absender und vernichten Sie diese E-Mail. Das unerlaubte Kopieren
sowie die unbefugte Weitergabe dieser E-Mail sind nicht gestattet.

This e-mail may contain confidential and/or privileged information. If you are not the intended recipient (or have received this e-mail in
error) please notify the sender immediately and destroy this e-mail. Any unauthorised copying, disclosure or distribution of the material
in this e-mail is strictly forbidden.


Am 02.10.2018 um 14:56 schrieb Geoff Boyle:

CDNG is a very good idea, at all the variations that are available in camera.

 

Last time I looked Cineform wasn’t an included option, perhaps I missed something in my haste to check out the new facilities in the latest firmware.

 

Now back to my original question…

 

What subject will show the effects of compression the most?

 

Bear in mind that it has to be simple, obvious and easily available.

 

I’ve been looking at the trees outside my window blowing in the wind. Lots of motion, fine detail, huge mix of light and dark, can be shot so that the entire image is changing every frame.

 

cheers
Geoff Boyle NSC
EU based cinematographer
+31 637155076

www.gboyle.nl

www.cinematography.net

 

 

From: cml-raw-log-hdr@... <cml-raw-log-hdr@...> On Behalf Of Colin Elves
Sent: 02 October 2018 10:08
To: cml-raw-log-hdr@...
Subject: Re: [cml-raw-log-hdr] BRAW tests

 

CDNG is the most obvious choice since that’s the other Raw format the UMP can encode directly. 

 

I guess the obvious choices are things that might cover: resolution, noise, colour resolution - and anything that might throw up debayering errors. Some people have suggested BRAW is a YCRCB codec as there is some ghosting in the red channel at high contrast edges.

 

Colin Elves

Director of Photography

UK Mob: +44 7886 849 073

German Mob: +49 157 78356204



 

 


On 2 Oct 2018, at 08:56, Jeff Kreines <jeff@...> wrote:

Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

 

Sent from iPhone. 


On Oct 2, 2018, at 1:39 AM, Geoff Boyle <geoff@...> wrote:

OK, I'm quiet now so have time to shoot tests.

What kind of things do you want to see in a tests? please bear in mind that I will be uploading 6 different versions of every test in BRAW so they will be big files.

Geoff Boyle NSC
Sysadmin
Zoetermeer

axel.mertes
 


Am 02.10.2018 um 08:56 schrieb Jeff Kreines:
Any way to do a comparison with that other excellent RAW format — Cineform?  That would be interesting. 

Jeff Kreines

Jeff, as far as I know you can't record in Cineform with the BMD cameras.
I could only transcode the BRAW result into a Cineform RAW. If I know the orientation of the Bayer pattern, I can strip out all interpolated pixels and re-create a Cineform RAW file.
However, we are then comparing the Cineform build-in RAW demosaic (and there are 5 options...) vs. BRAW demosaic. The rest will be the same.

What I don't get with BRAW is that it is not a codec I can use in post and until a master export as long as its treated as RAW. On the other hand, if its really a 444 codec inside, then that might change by at some point burning in color and using the same compression scheme plus a setting that is no longer the camera sensor specs and whitepoints etc. but those of my mastering color space like REC709, REC2020 or whatever comes to mind. No idea if that is on the roadmap or in any way possible.

Cineform 4444 is surely made for exactly this purpose.

I am preparing my codec test for this part, and one thing that I want to look at is iterative recompression. Here some codecs fail quickly, others hold up fantastic. I wan't to prove that with some sample content. Currently investigating which content might make sense for this, like downscaled 8K RED material or scanned film or simply BRAW files for instance. The point is that RED often demonstrates the camera with high noise detail imagery (like night sky large city skylines) vs. Geoff loving smoothed female skin tones... Thats two different shows and they are very different in post and in the context of compression etc.

Mit freundlichen Grüßen,
Best regards,

Axel Mertes

Workflow, IT, Research and Development
Geschäftsführer/CTO/Founder
Tel: +49 69 978837-20
eMail: Axel.Mertes@...

Magna Mana Production
Bildbearbeitung GmbH
Jakob-Latscha-Straße 3
60314 Frankfurt am Main
Germany
Tel: +49 69 978837-0
Fax: +49 69 978837-34
eMail: Info@...
Web: http://www.MagnaMana.com/

Diese E-Mail enthält vertrauliche und/oder rechtlich geschützte Informationen. Wenn Sie nicht der richtige Adressat sind oder diese
E-Mail irrtümlich erhalten haben, informieren Sie bitte sofort den Absender und vernichten Sie diese E-Mail. Das unerlaubte Kopieren
sowie die unbefugte Weitergabe dieser E-Mail sind nicht gestattet.

This e-mail may contain confidential and/or privileged information. If you are not the intended recipient (or have received this e-mail in
error) please notify the sender immediately and destroy this e-mail. Any unauthorised copying, disclosure or distribution of the material
in this e-mail is strictly forbidden.

_._,_._,_

Mark Weingartner, ASC
 

I use some of those circular pattern line pair charts (I like the ones Panavision Alga in Paris have) but so far my favorite has been a privett hedge that’s at panavision’s courtyard…

For intraframe compression lock-offs tell all, for interframe compression slow pans can break almost anything. Better yet when a breeze is blowing

Mark Weingartner, ASC
LA based DP/vfx

deanan@gmail.com
 

Sunlit Hedges or backlit leaves are great for getting high contrast / high detail that stresses a codec. 

Also having smooth  and detailed areas in shadow in various other parts of the frame helps see how the codec stomps on the shadow areas that most codecs affect strongly. That also helps see how the rate control does. 

Proves does rate control over the whole frame so a highly detailed part of a frame who the leave another part of the frame unnecessarily starved.  

Presumably braw is based off the same DCT IP as BMDs prides implementation. 

DeananDaSilva
Playa del Rey, Ca


>favorite has been a privett hedge that’s at panavision’s courtyard…


Mark Weingartner, ASC
LA based DP/vfx
_._,_._,_


Daniel Rozsnyó
 



On 10/02/2018 06:50 PM, axel.mertes wrote:
What might show off codec compression artefacts is usually high noise/frequency content, i.e. grass, leaves, gravel, windy water, moire-effect forcing "check patterns" or "pin stripe suit" silks. Then of course green sceen or blue screen with fine contour details like single hair, wavy curtains with embroidery, or fine grass in front of it. Even a not to close close up to really gray hair (should be available...) on a "hair thickness is not much more than 1-3 photosites" is often good to reveal color interpolation (demosaic) issues. Thats the stuff where we see it in post. The finer the detail, the more likely its hard to reproduce with compression. Each compression scheme tries to reduce high frequency noise - thats where the details go...

I would go for some sort of fabric, where we hit the resolution limit. But you still have to count in the effect of an OLPF.

It would be great to have the same scene with multiple settings (from totally uncompressed DNGs - please verify that the files are of constant size.. and they are not those JPEG DNG's which BM makes with 1:3/1:4 lossy codec to the various flavors of BRAW). Ideal setup would be to lock the camera and change settings remotely (if that is possible)

But then a very subtle and slow movement might help to find compression-breaking frames more easily.

What you have to count in, is the effect of "constant bitrate" vs "constant quality" setting difference:
 - on the constant quality, the scene contents does not matter, it will break same way whether it is half fabric and half sky
 - on a constant-bitrate setup, the fabric or other high-frequency content shall cover the whole picture, otherwise the portions where the "sky" is, will compress well and the saved bandwidth will be used for the other half, so your results will be not worth anything (unless the methodology includes the ratio of different scene contents)


Now comes the tricky part. There might be a codec which is a multi-pass one in order to get the best compression for the image regions while achieving a constant bit-rate per each frame frame, under i-frame setting. But for real-time capture applications, the encoder makers implement it with some cheats - among which an inter-frame dependency is the easiest one to avoid the need of full-frame recalculation (costs power and needs more powerful hardware).

So while you think that the codec is intra frame only, actually it is not and there might be a delay to adapt the compression. If the camera vendor refuses to tell how their implementation works, do not assume it is separate for each frame.


Resume:

    Constant quality = same quantizer = easy for hardware, but bit-rate will vary a lot

    Constant bit-rate:

        - per frame = needs different quantizer per portion of frame - in UHD BRAW there are 240 portions, in PR/PRR its even more finer), hardest to achieve, no single-pass perfect method exists - either you redo the whole frame, or cheat on the initial guess from previous frame. To get it in perfection the camera has to be "over-equipped" with the compression hardware, so that it can redo all the frames several times.

        - per group of frames = constant quantizer within a frame, that is adapted over longer period to meet the target bit-rate setting - that one will be definitely dependent on how rapidly the scene changes (e.g. revealing a fine fabric will show up perfectly sharp on 1st frame it appears, but its quality will degrade over consecutive frames in order to meet the bit-rate constraints)

        - (combination of above approaches)


PS: the same is true for e.g. AVC / HEVC - the codec specs definitely offer amazing features you can use for quality, but when you encode under real-time constraint (in camera), the quality is far from what it actually could be. This is especially true for codecs that are part of various SoC (system on chip) devices that target primarily low power (phones, etc.) - the easiest cuts employed are: limited motion vector search distance, the single size of macro-block, single quantizer over whole frame, etc.


Ing. Daniel Rozsnyo
camera developer
Prague, Czech Republic



Tobias Corts
 

So great to hear that you are doing this!!!
Skintones - RAW match to RAW
Dynamic Range.

Best,
Tobias Corts
www.tobiascorts.de
BavarianDOP

Aasulv Wolf Austad, FNF
 

Thanks for doing this!
If BRAW is indeed using DCT compression, try something that has 'broken' DCT compression in the past: Small face in the middle of a big white cyc. Given that it's compressing RAW, artifacts might of course show up in other places. I've also had major issues with film grain on DCT based compression of HDCAM and DVCAM, so high ISO with random noise might show things? Slow gradients are also an area where the shit hits the fan quickly. I've had some filters breaking DCT compression in the past, f.ex. Black Pro Mist used to 'break' DVCAM compression. That is probably too time consuming to test, though.
--
Aasulv 'Wolf' Austad, FNF
DP, Altadena, California
http://wolfaustad.com