Blackmagic Forum

What I would like to know is how is this done in 1080p? Also we know the active sensor size but is it larger to enable some sort of debayering in camera?

No, the sensor is native 1920 x 1080. The effective resolution will be slightly less, and less than the BMCC 2.5k scaled to 1080p.

For ProRes the image is debayered in camera, while for Cinema DNG raw it's not, obviously. If you process the RAW correctly it will likely be slightly sharper than filming in ProRes mode, due to more advanced debayering algorithms like those in Adobe Camera Raw.

Cheers Erik. I thought the sensor may have been slightly larger to allow in camera debayering.

Erik Swan wrote:No, the sensor is native 1920 x 1080. The effective resolution will be slightly less, and less than the BMCC 2.5k scaled to 1080p.

For ProRes the image is debayered in camera, while for Cinema DNG raw it's not, obviously. If you process the RAW correctly it will likely be slightly sharper than filming in ProRes mode, due to more advanced debayering algorithms like those in Adobe Camera Raw.

So the camera essentially line skips? If the sensor is native 1920x1080 then true debayering by downsize isn't happening.

ConstantProduction wrote:

Erik Swan wrote:No, the sensor is native 1920 x 1080. The effective resolution will be slightly less, and less than the BMCC 2.5k scaled to 1080p.

For ProRes the image is debayered in camera, while for Cinema DNG raw it's not, obviously. If you process the RAW correctly it will likely be slightly sharper than filming in ProRes mode, due to more advanced debayering algorithms like those in Adobe Camera Raw.

So the camera essentially line skips? If the sensor is native 1920x1080 then true debayering by downsize isn't happening.

No it doesn't skip lines.

sebasti wrote:

ConstantProduction wrote:

Erik Swan wrote:No, the sensor is native 1920 x 1080. The effective resolution will be slightly less, and less than the BMCC 2.5k scaled to 1080p.

For ProRes the image is debayered in camera, while for Cinema DNG raw it's not, obviously. If you process the RAW correctly it will likely be slightly sharper than filming in ProRes mode, due to more advanced debayering algorithms like those in Adobe Camera Raw.

So the camera essentially line skips? If the sensor is native 1920x1080 then true debayering by downsize isn't happening.

No it doesn't skip lines.

So how does it debayer in-camera?

It uses 1920 by 1080 photocells and de-bayers, what's the problem? This has nothing to do with line-skipping.

The true resolution is always lower than the number of photocells, the best de-bayering algorythms deliver about 75 to 80%.

The BMPCC is probably windowing on the sensor of the BMCC, not skipping any photocells.

ConstantProduction wrote:
So how does it debayer in-camera?

The Pocket cam sensor is basically the same as the BMCC, except it is cropped/windowed to be 1920x1080 active photosites instead of 2400x1350. The act of debayering either camera results in a video file, or image frame of either 1920x1080 pixels or 2400x1350. Note I am saying that the frame of video is that RESOLUTION, not that it has the resolving power of 1920x1080 or 2400x1350.

Since debayering is interpolation, each pixel of the debayered image contains information - some of which will have been derived from the debayering process rather than was captured at that specific photosite. So a green photosite has its Red and Blue values interpolated from the surrounding photosites. The result is indeed an image of the same resolution, albeit with interpolated chroma data.

Irrespective of this - the image is still 1920x1080 on the Pocket cam and 2400x1350 on the BMCC- in terms of image resolution.

The BMCC when scaled down to 1920x1080 using various methods of scaling such as bilinear scaling, yields an image of higher resolving power - closer to resolving 1080 lines (it won't actually resolve 1080 lines). The Pocket camera, will still produce a 1920x1080 image, but its resolving power will be much lower, lower than 1080 lines and lower than the BMCC.

I think the problem here is the confusion between image resolution and resolving power.

The reason that the Raw 1080p video of the pocket camera will be better quality in terms of sharpness than when filming prores - is because of the fact that different debayering methods produce different levels of quality. It is likely that the internal debayering method will not be as good as one done by say Resolve or Adobe Raw. Another reason is because the ProRes video is 4:2:2 - so some of the chroma data is discarded.

NOTE: Before anyone restarts the same old confusion between chroma sub-sampling and bayer sensors - they are not the same, the effect of compressing video by chroma sub-sampling is fairly uniform whereas the quality of chroma data from debayering is very varied and not at all uniform - they are not the same thing.


So in conclusion:

The resolution of the pocket camera is 1920x1080 - the resolving power can be worked out by smarter people than I, or with people who actually have the camera in their hands right now and a test chart.

After debayering - the image is still 1920x1080.

Thanks for that Tom - a very clear and concise explanation.

One has to understand, that with single chip sensor systems there is _always_ a difference between "official" resolution (= number of bayer pattern cells) and "measurable" resolution (= number of single pixels that can be resolved in the final image). There are some exotic sensors that can sample true color in one single pixel (actually it's three semi-translucent light cells behind each other), thus eliminating bayer pattern & debayering and reducing moiré, but as far as i know they've not yet made it into competitive cameras. An other solution is the use of three sensors as in the dying out 3ccd / 3mos cameras, but you simply couldn't built a device as tiny as the pocket cc with beam splitters and 3 sensor units...

Mac Jaeger wrote:One has to understand, that with single chip sensor systems there is _always_ a difference between "official" resolution (= number of bayer pattern cells) and "measurable" resolution (= number of single pixels that can be resolved in the final image).

Resolution vs resolving power.

resolution can be measured in pixels or photosites,

resolving power is usually related to the ability to resolve physical details above 50% MTF or before aliasing/artefacts occur.

So basically the pocket camera doesn't debayer RAW and so the image won't be as good as the BMCC?
So what will be the loss?

De-Bayering is simply the mathematical algorithm that re-assembles all the pixels in a bayer pattern sensor into an image.

All CMOS sensors have to have their image de-bayered or demosaiced. De-Bayering algorithms are constantly optimised and improved. Think of it as the chemistry that is used to *process* the image.

It's why you can have Adobe Camera RAW and Phase one load the same RAW image and get subtly different results. There are many imaging choices that can be made in the de-bayer process by those that write the de-mosiaic

Because the camera has to do it in real time and with the onboard CPU power available, it's generally considered to be better done in post, which is another reason why shooting RAW is good.

Some users have a workflow that uses ACR instead or resolve simple because they like the look of the ACR de-mosiac on black magic images.

What BMD have done is to improve the de-bayer algorithm. 99% of you won't even notice the differences. It will be subtle.

This new algorithm goes into the new Resolve 10 and will also trickle down int he cameras themselves with new firmware. This means ProRes and DNxHD encoding should benefit from the new de-bayer.

Line skipping has nothing to do with this.

This all happens independent of the sensor resolution which is a separate issue.

Many in *hollywood* consider Resolve to have the best debayer "engine" for all RAW formats, especially RED and ArriRaw. It's standard practice to grade using a redrocket (which great accelerates the de-mosiacing process) but then for the final render, to do it all in software and use Resolve's own de-bayer algorithm on the R3D files.

jb

Mark2929 wrote:So basically the pocket camera doesn't debayer RAW and so the image won't be as good as the BMCC?
So what will be the loss?

No, No one said that.

To film to a ProRes format means that the camera MUST be debayering internally. But there are so many different ways to debayer, some are simple and fast, others advanced and slow. The quality of the debayering method done internally for ProRes filming will almost certainly not be as advanced as one done by a pc or mac.

When the cameras is filming in its Raw mode, it wont be debayering internally, and will simply be saving bayer data - to be debayered later on by your computer.

In terms of how it sits with the BMCC - the image will be exactly the same as if you cropped the raw bmcc footage to 1920x1080. - not scaled down or line skipped or pixel binned, but cropped.

Tom wrote:
When the cameras is filming in its Raw mode, it wont be debayering internally, and will simply be saving bayer data - to be debayered later on by your computer.

In terms of how it sits with the BMCC - the image will be exactly the same as if you cropped the raw bmcc footage to 1920x1080. - not scaled down or line skipped or pixel binned, but cropped.

For RAW to debayer on the BMCC you have to scale down the 2400 to 1920
On the pocket camera you can't scale down as the sensor is 1920 so how can you debayer?

Unless it is a larger sensor in the pocket camera that debayers in camera to 1920?

Mark2929 wrote:
For RAW to debayer on the BMCC you have to scale down the 2400 to 1920

No.

You can debayer and stay at 2.5K if you want to. OR, you can scale down to 1920 from the 2.5k Source. Either way, you use the same *debayer* step. The debayer or demosiac is NOTHING to do with scaling the image.

Mark2929 wrote:On the pocket camera you can't scale down as the sensor is 1920 so how can you debayer?

Unless it is a larger sensor in the pocket camera that debayers in camera to 1920?

You debayer the 1920 sensor image and get a ProRes 1920 image that is something less than 1920 in terms of resolution. The file is 1920, but if you point it at a resolution chart it will never hit 1920.

It is true that because you don't have individual pixels for each colour, it's best practice to have a larger sensor size than what you're targeting so that you get a nicer image. This is why RED epic is a 5K camera, so they can claim a 4K image on screen.

So with the pocket one of the compromises is that it makes a nice 1920 ProRes file from a sensor that is 1920 in size. So no bayer sensor can actually resolve what it's pixel size is, and how close it can get to that number is based on how good the de-bayer algorithm is.

For those that say the pocket is just as good as the BMCC and somehow makes it redundant, this is the MAIN difference between these two cameras.

With the 2.5k sensor you'll get a "nicer" 1920 ProRes than from a straight 1920 sensor--> 1920 ProRes.

jb

So what will be the quality hit on the Pocket camera compared to the BMCC at 1080p?

Where can I find a great source on how bayer sensors and the whole debayering process works I have read some stuff but want to know more... Any good resources out there?

Justin Phillips wrote:Where can I find a great source on how bayer sensors and the whole debayering process works I have read some stuff but want to know more... Any good resources out there?

http://www.red.com/learn/red-101/bayer-sensor-strategy

http://blog.elphel.com/2010/11/zoom-in-now-enhance/ (where if you scroll down, you can see comparisons between different debayering methods -
-
)

and of course, wikipedia: https://en.wikipedia.org/wiki/Bayer_filter

Thanks a lot Tom!

Mark2929 wrote:So what will be the quality hit on the Pocket camera compared to the BMCC at 1080p?

How to measure "quality hit"? The 2k5 camera has 25% more pixels, bayer cells or debayered, doesn't matter. In theory this means that you get 25% more resolution, but under real world conditions this only means you get a little less moiré and a little more details. Yet just "how little" can not be put into numbers. Probably less then what you loose when you render your final work out to BluRay or - even worse - internet codecs.

So.. In 1080p no real noticeable difference on a 40' screen.

Mark2929 wrote:So.. In 1080p no real noticeable difference on a 40' screen.

Those terms are subjective.


The only real answers that can be given are:

1: The Pocket cameras has a 1920x1080 image sensor - no line skipping or pixel binning or scaling.
2: The resolution in Raw mode is lower than that of the BMCC.
3: When you scale the BMCC footage from 2.5k RAW to 1080p, it will appear to resolve more detail than the Pocket Camera.

Mac Jaeger wrote:... There are some exotic sensors that can sample true color in one single pixel (actually it's three semi-translucent light cells behind each other), thus eliminating bayer pattern & debayering and reducing moiré, but as far as i know they've not yet made it into competitive cameras...

An example of this technique may be the foveon sensor owned and used by Sigma which exploits the behaviour of light travelling through silicon so the first light captured is blue, then deeper it is green, and at the bottom is red. Currently red is the biggest challenge and the math is intensive to summarize neat values for the three colours. Will be amazing when it is ready for prime time. It uses CMOS but differently than the bayer filter concept.

Rick Lang
Sent from my iPad using Tapatalk HD

John Brawley wrote:With the 2.5k sensor you'll get a "nicer" 1920 ProRes than from a straight 1920 sensor--> 1920 ProRes.

jb

Just want make sure regarding to this point, if my final output just a 1920 ProRes (for web purposely) so it will be better if we shoot in RAW then compressed it to 1920 ProRes. Am i missing something?

Rudy Satria wrote:

John Brawley wrote:With the 2.5k sensor you'll get a "nicer" 1920 ProRes than from a straight 1920 sensor--> 1920 ProRes.

jb

Just want make sure regarding to this point, if my final output just a 1920 ProRes (for web purposely) so it will be better if we shoot in RAW then compressed it to 1920 ProRes. Am i missing something?

You will be very hard pressed to actually *see* the difference, but generally, Resolve will give you a better debayer than what the in-camera debayer will be. But you're more likely to detect lens imperfections before you'll see debayering artefacts. Please don't get hung up about the *improved* debayer. It's a small part of the process.

jb

Rudy Satria wrote:

John Brawley wrote:With the 2.5k sensor you'll get a "nicer" 1920 ProRes than from a straight 1920 sensor--> 1920 ProRes.

jb

Just want make sure regarding to this point, if my final output just a 1920 ProRes (for web purposely) so it will be better if we shoot in RAW then compressed it to 1920 ProRes. Am i missing something?

Sorry for digging this up so many times this week in so many posts, but here is an example I made to show the difference:

http://tommajerski.com/publicimages/Raw2_1.4.1.jpg

Slight difference in both resolving power and also aliasing.

Tom wrote:Sorry for digging this up so many times this week in so many posts, but here is an example I made to show the difference:

http://tommajerski.com/publicimages/Raw2_1.4.1.jpg

Slight difference in both resolving power and also aliasing.

Is that zoomed in at all Tom?

Mark2929 wrote:

Tom wrote:Sorry for digging this up so many times this week in so many posts, but here is an example I made to show the difference:

http://tommajerski.com/publicimages/Raw2_1.4.1.jpg

Slight difference in both resolving power and also aliasing.

Is that zoomed in at all Tom?

no, it is a 100% crop

Tom wrote:

Rudy Satria wrote:…here is an example I made to show the difference:

http://tommajerski.com/publicimages/Raw2_1.4.1.jpg

Slight difference in both resolving power and also aliasing.

Although the difference may become less significant with future firmware. I believe a firmware update has already been announced with an improved deBayering algorithm. I suspect this may have been developed with the Pocket Camera mind, because that does not have the down-scale to mask deBayering artefacts.

ARRI did a similar thing with the ALEXA, introducing an in-camera deBayering algorithm previously only available in post with SUP 7.0

John Brawley wrote:You will be very hard pressed to actually *see* the difference, but generally, Resolve will give you a better debayer than what the in-camera debayer will be. But you're more likely to detect lens imperfections before you'll see debayering artefacts. Please don't get hung up about the *improved* debayer. It's a small part of the process.

jb

Thanks John. It seems very clear now

Tom wrote:Sorry for digging this up so many times this week in so many posts, but here is an example I made to show the difference:

http://tommajerski.com/publicimages/Raw2_1.4.1.jpg

Slight difference in both resolving power and also aliasing.

Thanks Tom for showing the example. Actually im a bit confuse to define Aliasing and Moire. Do you have any link or post showing side by side which one is Aliasing and which one is moire so i can clearly see the different? Thanks

Rudy Satria wrote:
Thanks Tom for showing the example. Actually im a bit confuse to define Aliasing and Moire. Do you have any link or post showing side by side which one is Aliasing and which one is moire so i can clearly see the different? Thanks

Aliasing = artefacts caused by the effects of sampling
Moire = a type of aliasing caused by spatial sampling - visually manifested as strange patterns not physically present in the real world as it is being filmed.

The colour aliasing present within my example image is most likely caused by the detail being so fine that the interpolation process during debayering has resulted in chroma aliasing.