8 bit vs 12 bit

[Steve Phillipps]

As well as being a way to keep costs down (at the possible expense of picture quality) is it correct that lower spec processing also lowers power consumption considerably? I think the FS700 can run on the tiny Sony batteries for hours like the EX1 etc. whereas Scarlet is pretty thirsty.
It does seem to me that the 8 bit processing would be a very good reason why the FS700 is a lot cheaper than a Scarlet even though it can do 240fps in full HD. The footage I've seen from the FS700 in general has looked way better than a lot of Scarlet footage I've seen on the web - is it possible that the 8 bit limitations would only show themsleves when on a proper display?
Steve

 

[Scott Crawley]

As well as being a way to keep costs down (at the possible expense of picture quality) is it correct that lower spec processing also lowers power consumption considerably? I think the FS700 can run on the tiny Sony batteries for hours like the EX1 etc. whereas Scarlet is pretty thirsty.
It does seem to me that the 8 bit processing would be a very good reason why the FS700 is a lot cheaper than a Scarlet even though it can do 240fps in full HD.

That may be the case. It is certainly possible.

The footage I've seen from the FS700 in general has looked way better than a lot of Scarlet footage I've seen on the web - is it possible that the 8 bit limitations would only show themsleves when on a proper display?

I don't know what footage you are referring to but there are a myriad of possible reasons for the differences you are seeing, and the vast majority of those reasons are related to the shooting and post skills of the folks producing the footage. It is not your display that is responsible, no matter how crappy it may or may not be. ;-)

 

[Steve Phillipps]

I think you're right - with shoooting for sure, but most definitely with post, no doubt partly due to the nature of RAW workflow, I think it's the same with stills - if you really know what you're doing you can change an image hugely for the better.
I think the other thing some Scarlet users may be doing is exagerating the film look, making it to flat and shallow depth of field - often this can just end up looking muddy and soft. On the contrary some of the Sony stuff looks bright and sharp, which is good, but undoubtedly more videoy which is what a lot of people are wanting to avoid - but I think there's a happy medium there somewhere.
I also get the feeling that there's a huge gulf between downsampled 4k and 2k originated material - on Scarlet and Epic. On Epic it's not a problem as you can go to 120fps at 4k, but on Scarlet if you need 4k to look really great then you're limited to 30fps. Lots of ifs and buts though!
Steve

 

[Hector Berrebi]

can anyone who mentioned it validate that REDCODE RAW actually writes 16 bit?

i was under the impression it was 16 bit processing, and writes 12 bits. which is stil plenty...

it is important to remember, that 12 bits are not supported by most apps, most codecs and most workflows,
12 bit monitors are not something you "just buy" or even simply rent. and if i'm correct, then (at least until we can order the REDRAY projector) expensive high-end digital projection is the main option to see 12 bits.

monitoring/viewing 16 bits? i don't think anytime soon

of course, you can work in 16 bit and even 32 bit environments, and use EXR or TIFF as your file formats to conserve the color data,
its not simple. and there are more ways to lose color depth than maintain it,

all this because, like the FS700, the great majority of the video world is still, and probably will stay for a while, 8 bits.

even 10 bits are considered luxury in most workflows.

that's why 12 bit R3Ds are so amazing

i'm not even sure if the 12 bits of color data that R3D files produce are not more impressive to me than the 4K resolution.



hector

 

[Lee Saxon]

FS700, which in my opinion is nothing more than an overpriced toy, is not only 8 bit its 4:2:0 too...

Have to make the FS700 8-bit 4:2:0 because otherwise no one would buy the F65, amirite? Yeah, Sony is full of shit. The Nikon D800 outputs 4:2:2, for God's sake.

 

[paul ross jones]

im a stills guy who grades shots a lot. every now and then I'm doing a composite that requires a stock shot- i.e. a sky that we weren't able to get. so we get a jpeg off getty or someplace like that.

so we mix in the jpeg along with 16bit medium format files and its pretty obvious the fragileness of the jpeg files. but if we convert the jpegs into 16 bit tiffs the files improve significantly. Photoshop makes up the shades in-between the existing ones.
Sure the files still aren't as good as the 16bit medium format files, but they are way better than the original jpeg files.

I saw the same thing when grading some 5dmk3 footage recently- when the footage was converted to 16 bit in the grading suite the footage improved significantly. the shots didn't get anymore lattertude- i.e., when the top is blown out the was no fixing it, but the graduations in the tones that were within range could be twisted far more.

This is important for me, as I really want to get a camera that does a decent slow motion, so the fs700 is very appealing. the other day i shot some test footage with the fs700 onto the internal card and into a samurai at the same time. I'm about to take this down to the my colourist to see his opinion on it. The samurai interpolates (it this is the right term..) the 8 bit 4.2.2 to 10bit 4.2.2. I am hoping there is a significant improvement with the footage like i have seen with 5d footage in the grading suite.

sure this won't be as good as "close to raw" that an epic/scarlet does, but for my budget, the extra production value of slow motion is more important to me than a bit more latitude and a few more shades of tone/colour.

The footage i see on the web from the fs700 does look very "video" to me, but often the red footage does as well... i guess its internet compression or the way its been shot. I'm going to keep testing and see if i can get the fs700 to look better, if it can't, i guess I need to invest more into a camera.

paul

 

[Dave Blackham]

The key take away is you need at least 10-bit depth (i.e. precision) in your recording to avoid visible banding in smooth graduated backgrounds. Higher bit depth is important if you intend to do extensive color correction, so that you can maintain that minimum precision even after pushing and pulling the data around.

Which is why RED has always done at least 12-bit precision recording. Recording RAW v's YCbCr is not relevant to the above...

Agreed this is important. I hope my understanding is correct, Im not sure 10bit is now good enough. Strictly 10bit yeilds 1024 colours per colour channel, so for Standard definition its OK with a canvas size of 720 x 576 in PAL land, as it has more colours than oriental pixels per line to render a perfect gray or coloured ramp. but it isn't good for HD at 1920 x 1080 which would need 11bit minimum at 2048 colours per colour channel. So for 5k you would need more than 5k colours per colour channel. 13 bit but in practice would be impractical, I guess arguably you could make 14bit work but more likely to use 16 bit as computers use 8 bit words and its easier ti implement and also its a widely used standard. Also as always things improve 6k and 8k and so on

 

[Hector Berrebi]

im a stills guy who grades shots a lot. every now and then I'm doing a composite that requires a stock shot- i.e. a sky that we weren't able to get. so we get a jpeg off getty or someplace like that.

so we mix in the jpeg along with 16bit medium format files and its pretty obvious the fragileness of the jpeg files. but if we convert the jpegs into 16 bit tiffs the files improve significantly. Photoshop makes up the shades in-between the existing ones.
Sure the files still aren't as good as the 16bit medium format files, but they are way better than the original jpeg files.

I saw the same thing when grading some 5dmk3 footage recently- when the footage was converted to 16 bit in the grading suite the footage improved significantly. the shots didn't get anymore lattertude- i.e., when the top is blown out the was no fixing it, but the graduations in the tones that were within range could be twisted far more.

This is important for me, as I really want to get a camera that does a decent slow motion, so the fs700 is very appealing. the other day i shot some test footage with the fs700 onto the internal card and into a samurai at the same time. I'm about to take this down to the my colourist to see his opinion on it. The samurai interpolates (it this is the right term..) the 8 bit 4.2.2 to 10bit 4.2.2. I am hoping there is a significant improvement with the footage like i have seen with 5d footage in the grading suite.

sure this won't be as good as "close to raw" that an epic/scarlet does, but for my budget, the extra production value of slow motion is more important to me than a bit more latitude and a few more shades of tone/colour.

The footage i see on the web from the fs700 does look very "video" to me, but often the red footage does as well... i guess its internet compression or the way its been shot. I'm going to keep testing and see if i can get the fs700 to look better, if it can't, i guess I need to invest more into a camera.

paul

paul. you cannot fight math
recording an 8 bit signal to a 10 bit codec will result in 2 empty bits (values at zero)

what you will get from an external recorder is 4:2:2 (50% more color information) and about 10X more Mbs

I think the 5k example you give, is mainly due to the 16 bit environment and superior grading tools, as well as the grader's skills, not necessarily an improvement in the file's colors.

now you simply have 16.7 million colors, arranged over a range of many bilions (do 65,000x65,000x65,000) ...


hector

 

[Noah Yuan-Vogel]

Have to make the FS700 8-bit 4:2:0 because otherwise no one would buy the F65, amirite? Yeah, Sony is full of shit. The Nikon D800 outputs 4:2:2, for God's sake.

Nikon D800 does not record 4:2:2 internally does it? FS700 outputs 4:4:4 over HDMI, but good luck recording more than 4:2:2 from it.

 

[Hector Berrebi]

Nikon D800 does not record 4:2:2 internally does it? FS700 outputs 4:4:4 over HDMI, but good luck recording more than 4:2:2 from it.

you'd need a recorder that has an internal 4:4:4 codec... and takes HDMI 1.4
i don't think the make them yet
maybe on a computer with a capture card, to ProRes 4x4...

 

[paul ross jones]

paul. you cannot fight math
recording an 8 bit signal to a 10 bit codec will result in 2 empty bits (values at zero)

what you will get from an external recorder is 4:2:2 (50% more color information) and about 10X more Mbs

I think the 5k example you give, is mainly due to the 16 bit environment and superior grading tools, as well as the grader's skills, not necessarily an improvement in the file's colors.

now you simply have 16.7 million colors, arranged over a range of many bilions (do 65,000x65,000x65,000) ...


hector

Hi Hector, will the external HD make a big difference given it goes straight to a 4.2.2 prores (or DNxHD) when the hard drive - but padded out to 10 bit?
I understood when a program like photoshop turned an 8bit file into a 16bit file, it "made up" the shades that are missing, but seems to work- you can see it when you grade a sky for example.
I assumed that the samurai was doing the same (the literature seems to say this..).

Everyone I talk to at the post house say how they don't like AVCHD, so i was hoping the external HD would improve things a lot..

Cheers Paul

 

[Hector Berrebi]

Everyone I talk to at the post house say how they don't like AVCHD, so i was hoping the external HD would improve things a lot..

Cheers Paul

it will make a big difference.
4:2:2 is considerably superior to 4:2:0, especially in the red channel, somewhat in the blue.
220 Mbs is 10 times more data than 24 Mbs ,and its iFrame, not longGOP.
you will see clear differences in moving small detail,
like a girl shaking her head and the (well lit) hair moves completely from one frame to another, or dancers/models with fine patterns on their dresses
low bit-rate + interframe compressions produce blocking in these situations, ProRes works better. it also means better tracking, and better keying.
and generally better grading

it will always lack something in its color pallet, something most viewers won't know even how to describe, 8 bit is not 10 bit (when properly treated).
but it will be far better than internal recording.

this is F3, so 10 bit recording to KiPro, but a good example of this phenomenon
http://prepostconsulting.com/en/english-external-recorders-part-3-kipro-vs-sony-f3/

hector

 

[Scott Crawley]

4:2:2 is considerably superior to 4:2:0, especially in the red channel, somewhat in the blue.

4:2:2 (Y:Cr:Cb) indicates equal sampling of red, and superior sampling of blue when compared to 4:2:0. Would you mind clarifying what you mean?

 

[Hector Berrebi]

4:2:2 (Y:Cr:Cb) indicates equal sampling of red, and superior sampling of blue when compared to 4:2:0. Would you mind clarifying what you mean?

hey Scott
its a bit complex, but 4:2:0 does not mean no sampling of blue. logically it is more like 4:1:1 and what it means is that about 75% fo chroma information is gone.
4:2:0 uses a smarter matrix in the way it samples the Cr and Cb channels for better reconstruction.

blue channel is a bit louder in video (i think because of sensors... not sure)
even in R3D files the blue channel is often quite noisier than the others

when going 4:2:2 you will mainly notice an improvement in the red channel. and a "somewhat" improvement in the blue channel, it still is a better blue channel than in the 4:2:0.

I'm not really a video technician and probably a better explanation can be made.
However, i'v seen it happen in several parallel test shoot i did with camera codec at 4:2:0 and external recorder at 4:2:2.

 

[Scott Crawley]

hey Scott
its a bit complex, but 4:2:0 does not mean no sampling of blue. logically it is more like 4:1:1 and what it means is that about 75% fo chroma information is gone.
4:2:0 uses a smarter matrix in the way it samples the Cr and Cb channels for better reconstruction.

blue channel is a bit louder in video (i think because of sensors... not sure)
even in R3D files the blue channel is often quite noisier than the others

when going 4:2:2 you will mainly notice an improvement in the red channel. and a "somewhat" improvement in the blue channel, it still is a better blue channel than in the 4:2:0.

I'm not really a video technician and probably a better explanation can be made.
However, i'v seen it happen in several parallel test shoot i did with camera codec at 4:2:0 and external recorder at 4:2:2.

It is a ratio. 4:2:0 means that for every 4 sample of luma taken, 2 red samples and less than one sample of blue are taken. True, it does not mean no blue. Likewise 4:2:2 means for every 4 luma samples, 2 red and 2 blue are taken. So in terms of chroma sub-sampling a 4:2:0 and 4:2:2 sample are essentially equal in red.

If when you state:

it will make a big difference.
4:2:2 is considerably superior to 4:2:0, especially in the red channel, somewhat in the blue.
220 Mbs is 10 times more data than 24 Mbs ,and its iFrame, not longGOP.
you will see clear differences in moving small detail,
like a girl shaking her head and the (well lit) hair moves completely from one frame to another, or dancers/models with fine patterns on their dresses
low bit-rate + interframe compressions produce blocking in these situations, ProRes works better. it also means better tracking, and better keying.
and generally better grading

it will always lack something in its color pallet, something most viewers won't know even how to describe, 8 bit is not 10 bit (when properly treated).
but it will be far better than internal recording.

this is F3, so 10 bit recording to KiPro, but a good example of this phenomenon
http://prepostconsulting.com/en/english-external-recorders-part-3-kipro-vs-sony-f3/

hector

you mean that you are seeing substantially better Red Channels in 4:2:2 ProRes files than the 4:2:0 AVCHD files taken from the same camera, I would suggest that those are not due to the chroma sampling as you seem to suggest, but are due to some other compromise the codec is making in it's compression.

(Edit) Getting back on topic, in the given scenario the difference is more likely due to the fact that the camera native AVCHD files are 8-bit and the Ki Pro ProRes files are interpolated to 10-bit. It's not about the relative number of samples, but rather the quality of those samples.

When the Ki Pro does it's interpolation, do we know for certain that it is simply adding two empty bits as Hector states, or is it creating interpolated info in those bits, as in Paul's Photoshop example?

 

[Hector Berrebi]

It is a ratio. 4:2:0 means that for every 4 sample of luma taken, 2 red samples and less than one sample of blue are taken. True, it does not mean no blue. Likewise 4:2:2 means for every 4 luma samples, 2 red and 2 blue are taken. So in terms of chroma sub-sampling a 4:2:0 and 4:2:2 sample are essentially equal in red.

from wikipedia (on chroma subsampling)
"4:2:0

In 4:2:0, the horizontal sampling is doubled compared to 4:1:1, but as the Cb and Cr channels are only sampled on each alternate line in this scheme, the vertical resolution is halved. The data rate is thus the same"
http://en.wikipedia.org/wiki/Chroma_subsampling

there are no 2 samples of red for every 4 of luma in 4:2:0, just a different matrix resulting in same data as 4:1:1, its just, very confusingly, called this way.

 

[Scott Crawley]

So it's not a ratio at all, it's a description of a matrix. I must admit, any time I tried to make sense of the graphics my eyes glazed over.

An actual ratio would have to account for black and white samples of luma, as well as red and blue, right? So to express it as a ratio it would have to measure two lines at a time, and be something like 8:8:2:2 or 4:4:1:1 (b:w:r:b)?

Sorry if this seems way off topic, but it is not so far off as you might think.