When is 4:4:4 not really 4:4:4 (Alister Chapman)

exactly as stated, its nowhere near the price of the f35 or f9000.

though the less is done to the signal the better.

Lexicon are you waiting to get your f3? if youre in NY and you'd like to check it out i will have one by the weekend.
 
Not really true. The fact that there are more green pixels than red/blue doesn't mean red/blue channels are chroma subsampled.
You need to compare the actual number of sensor photocells for given color to the output raster size. If the number of, say, red pixels equals to the number of pixels in the output raster, there's no subsampling at all. In such case the fact that it's the Bayer sensor (combining photocells for all three colors) only only affects the fill factor (which has it's disadvantages). Provided there's sufficient number of photocells true 4:4:4 can be get out of it. And since there's twice as much green your green channel can be oversampled, in fact.
Sony says F3 sensor has 3.36 megapixels effective resolution. We don't really know what 'effective resolution' means in case of Q67 sensor layout, but there's probably more photocells than 1920x1080 raster (~2mpx).
 
Sony removed the 3.36 megapixel effective resolution comment from the marketing materials. May have notbteen accurate at all, plus people aren't agreeing on what "effective" resolution may have referred to specifically anyway.
 
the whole article touches on a point of the inherent drawback of CMOS sensors. PERIOD. not a single CMOS sensor isn't prone to lose of "color" information due to how debayering works. plain and simple. that is what CCD does for you, it gives you the highest color rendition possible, but there are a LOT of drawbacks of CCD too. also there are literally dozens upon dozens of debayering algorithms out there, each one providing different quality results, some giving a lot more color information back than others. it's not black and white on how you gain all the information from your pixels, there are a lot of factors.

compression, pixel size, pixel depth, pixel shape. how it's debayered, how much effective resolution from the debayer are you getting. it isn't so cookie cutter. I was grading phantom cine footage this past weekend, and that's all from a CMOS sensor. your jaws would drop to the floor seeing how much color information is there.

when dual link comes out, and I hook a cinedeck to it, the proof will be in the pudding as they say.
 
which you need for CMOS sensors, because they tend to be single chip with a filter array.

CCD doesn't have that. f3 is cmos, f35 is CCD, those are the major gains in f35 of color rendition. for a CCD camera you gain your full bit depth in your RGB per pixel, if it's 10 bit, it's 10 bit per channel. making it 30 bit per pixel. that's where you get such great color from. when it comes to CMOS, and a filter array, and debayering algorithm, there are a lot of concerns on the color rendition. which are well warranted, and what i'm trying to say is, there isn't one flavor to the recipe when it comes to CMOS, single chip, filter array. there are dozens of ways to get color information, and there is no telling how close to a f35 color rendition the f3 is capable of until dual link comes out, and you hook up a off board recorder that can give you little to no compression at it's highest quality. then, and only then can you find out how good it is.
 
xmephestox said:
which you need for CMOS sensors, because they tend to be single chip with a filter array.

CCD doesn't have that. f3 is cmos, f35 is CCD, those are the major gains in f35 of color rendition. for a CCD camera you gain your full bit depth in your RGB per pixel, if it's 10 bit, it's 10 bit per channel. making it 30 bit per pixel. that's where you get such great color from. when it comes to CMOS, and a filter array, and debayering algorithm, there are a lot of concerns on the color rendition. which are well warranted, and what i'm trying to say is, there isn't one flavor to the recipe when it comes to CMOS, single chip, filter array. there are dozens of ways to get color information, and there is no telling how close to a f35 color rendition the f3 is capable of until dual link comes out, and you hook up a off board recorder that can give you little to no compression at it's highest quality. then, and only then can you find out how good it is.
Click to expand...

Uhh, what? the f35 uses one large bayer pattern sensor, exactly like the f3. It just happens to be a CCD. The fact that one is CMOS and the other CCD has nothing to do with any of this. They both operate in the exact same manor, using the exact same debayering method. The f35 just (supposedly) has more photosites in it's single sensor. I say supposedly, because nobody knows the photosite count in the f3's sensor, which is why I really wish people like Alister would quit this until we have solid information. It's nothing more than speculation, typed up like it's solid, end of the world information. Even if it has a smaller photosite count than theoretically needed to get 444 rgb output at full raster 1080, the interpolated samples from the bayer layout interpolated in the raw, 14-bit processing space will be much, much more solid data to comprise 444 than just recording 422 and interpolating to 444 in software. The fact that alister compares this to "putting 8 bits in a 10 bit stream and calling it 10 bits" is laughable, after his post on s-log and now this I'm really starting to wonder where he got his technical background. Also, speaking of bit depth, why would you bring that into the conversation? A CMOS sensor outputs raw analog voltage, just like a CCD, which is processed in a 14-bit space after being A/D'ed. When processing is done it is dithered down to the 10-bit SDI output. Just like on an f35. Their aren't "dozens of ways to get color information" and "lots of concerns on the color rendition", a single-sensor bayer pattern setup is a very standardized methodology, proven time and time again. Also, nobody knows if this sensor is even bayer in the first place, or sony's Q67. So once again, threads like these are all speculation.

Edit: just kidding, I was thinking of the D-20. The imaging sensor in an f35 is an rgb striped CCD, not bayer



BobbyMurcerFan said:
It's actually "worse" than that. If you can't get true 4:4:4 off the chip, then you can't get true 4:2:2 either. Even true 4:2:0 might not be possible b/c you never have a full array of just green.

In practice, it seems that oversampling on the chip and demosaicing algorithms do a good job of reconstructing the image.
Click to expand...

Uh, what? it is easily going to have a high enough photosite count to get a non-interpolated 4:2:2 image after debayer, in reality I'm sure it will have a high enough photosite count to properly debayer to 444. But like I said earlier, nobody knows what that count is so threads like this are silly. Even if the count is anywhere near 3.43mp as sony originally stated, it would have more than a full array of green to sample from.
 
Last edited:
Fohdeesha said:
Uhh, what? the f35 uses one large bayer pattern sensor, exactly like the f3. It just happens to be a CCD. The fact that one is CMOS and the other CCD has nothing to do with any of this. They both operate in the exact same manor, using the exact same debayering method. The f35 just (supposedly) has more photosites in it's single sensor. I say supposedly, because nobody knows the photosite count in the f3's sensor, which is why I really wish people like Alister would quit this until we have solid information. It's nothing more than speculation, typed up like it's solid, end of the world information. Even if it has a smaller photosite count than theoretically needed to get 444 rgb output at full raster 1080, the interpolated samples from the bayer layout interpolated in the raw, 14-bit processing space will be much, much more solid data to comprise 444 than just recording 422 and interpolating to 444 in software. The fact that alister compares this to "putting 8 bits in a 10 bit stream and calling it 10 bits" is laughable, after his post on s-log and now this I'm really starting to wonder where he got his technical background. Also, speaking of bit depth, why would you bring that into the conversation? A CMOS sensor outputs raw analog voltage, just like a CCD, which is processed in a 14-bit space after being A/D'ed. When processing is done it is dithered down to the 10-bit SDI output. Just like on an f35. Their aren't "dozens of ways to get color information" and "lots of concerns on the color rendition", a single-sensor bayer pattern setup is a very standardized methodology, proven time and time again. After reading Alisters post, it's pretty clear he does not fully understand how bayer sensors comprise data. Also, nobody knows if this sensor is even bayer in the first place, or sony's Q67. So once again, threads like these are all speculation.





Uh, what? it is easily going to have a high enough photosite count to get a non-interpolated 4:2:2 image after debayer, in reality I'm sure it will have a high enough photosite count to properly debayer to 444. But like I said earlier, nobody knows what that count is so threads like this are silly. Even if the count is anywhere near 3.43mp as sony originally stated, it would have more than a full array of green to sample from.
Click to expand...

Actually, the F35 is not Bayer, it's vertical stripe. It has enough pixels that there is a R, G and B sample for each portion if you cut up the imager into 1920 by 1080.

As for my comment, I don't know what you have an issue with. If you accept that a "X" camera's sensor can't provide enough info., due to its Bayer mask, to create true 4:4:4, then it can't create true 4:2:2 either.
 
That's the thing, the whole point of my post is that we do not know if it can provide enough info or not, because nobody knows its photosite count. We also do not know if it is even bayer in the first place, or rotated Q67 as sony promised to roll out some time ago. Either way, it will definitely have a full array of at least green for proper debayer, or Q67 processing. And Interesting! I was under the impression it was a Super35 bayered CCD with a smaller pixel pitch
 
Last edited:
TimurCivan said:
Lexicon are you waiting to get your f3? if youre in NY and you'd like to check it out i will have one by the weekend.
Click to expand...

I am currently in Chile, but, thanks TimurCivan for the that, It's very kind of you. Plans to be in NY around April.
 
and there are other considerations to look into. f35 comes from a much high resolution sensor. somewhere in the 5k-5.5k range. If it's near identical to the genesis sensor, than it should be in the 5.5k range. sony f3 is coming from around a 3.3 megapixel sensor and it debayers. the pixel size on it is claimed to be MUCh larger than the f35, hence the high sensitivity of the sensor. there are a lot of things to account for, and I'd like to think the f3 is using a better CFA method. there are a lot of things to consider here, and all i'm trying to say is it's not so cookie cutter. we gotta get the camera in our hands and see what it can do. only way we can make critical analysis, not just speculation. debayering whichever way you do it has it's difference and will provide different results.
 
when dual link is available with s-log in the near future. I'm gonna personally try to get the highest quality footage I can from it off a cinedeck. either uncompressed or in cineform. I have phantom and r3d to compare with, and I'm gonna try to get something off a f35 possibly even an Alexa. It's gonna go in my Da Vinci, and I'm gonna see for myself how close the f3 compares to the top dogs. I plan to actually make it unscientific, just to show what I do in the grade day in and day out for clients. contrast expansion, loads of power windows, hsl qualifiers, the works. that's the only way in my eyes to see if the camera is up to snuff. I can already say working with Phantom 14 bit uncompressed 1080p is my highest standard right now, with what I have worked with. noise is practically non existent, 13 stops of DR, huge amount in the low end especially( even better than the alexa and red) and I can move color around like it was nothing.

i think that's gonna be the most interesting way to see a test. I'm tired of charts and graphs. it's always great to say this camera can do this and do that, I wanna see how it LOOKS and how easy it is to WORK with.
 
Do you or anyone know or have heard anything if the f3's sensor is Q67 or not? It would make much more sense given the numbers sony has let out. Also, does anyone know the exact width of the sensor (super35=inches?) With that data plus it's photosites pixel pitch (12 microns) it wouldn't be hard to get a rough photosite count at least on the horizontal plane. Assuming the sensor is laid out in such a way. Then deducing it's bayer capabilities wouldn't be very hard.
 
Well uh oh, assuming the provided information of a 12 micron photosite pitch is correct, and the measurements given in the brochure, they would only be able to fit 1966 photosites horizontally and 1108 vertically. That's assuming the photosites cover the entire sensor, are back to back, and square. And yes, that is a lot of assuming. From that data, the best case scenario is a 2.1MP sensor, photosite count not pixel count, before debayer. So something's gotta be wrong here! They must be providing photosite pitch as measuring a group of photosites required for one pixel output. So that still does not tell us how many physical photosites they are using to deduce pixels.
 
Last edited:
Fohdeesha said:
That's the thing, the whole point of my post is that we do not know if it can provide enough info or not, because nobody knows its photosite count. We also do not know if it is even bayer in the first place, or rotated Q67 as sony promised to roll out some time ago. Either way, it will definitely have a full array of at least green for proper debayer, or Q67 processing. And Interesting! I was under the impression it was a Super35 bayered CCD with a smaller pixel pitch
Click to expand...

Fine, but I'm not talking about the F3 specifically. I'm making a more general comment on the conversion from Bayer to output. Simply that everyone seems to be under the impression that a Bayer to 4:2:2 is okay but Bayer to 4:4:4 is not. In reality, if it cannot provide true 4:4:4 then it cannot provide true 4:2:2 either. Even 4:2:0 will have interpolation going on. That's the nature of a Bayer pattern.

Now how bad this effects the image depends on a lot of factors. But I don't see why the interoplation used to go 4:2:2 is okay while going to 4:4:4 is not. It's the same interpolation, just applied to H and V resolution instead of just V.


P.S. The F3 is not Q67.
 
Last edited:
How a Bayer pattern sensor is interpolated to produce a color image is simply not directly comparable to how an encoded video format stores that color information. It is apples and oranges. They are two entirely separate processes. Bayer pattern sensors are the overwhelming choice for digital still cameras because they can produce color gamuts for fine quality printing processes that far exceed 10 bit video, 4:4:4 or not.
 
Three meanings of the term 'chroma subsampling'

Three meanings of the term 'chroma subsampling'

After reading Alister's post I concluded that we should distinguish between three different and independent senses of the term 'chroma subsampling':

1. In one case we refer to the information that the sensor is capturing due to its color filter array (CFA). In the case of a Bayer CFA, the sensor is indeed 'chroma subsampling' the incoming information, since it is sampling two more times the green frequencies than the red and blue ones. If you have three CCDs (one dedicated for each color) or a single sensor with a strip type of CFA (like in the F35), those will be cases in which you are not 'chroma subsampling' in this sense of the term.

2. Another sense of the term 'chroma subsampling' has to do with the comparison between the actual number of sensor photocells for a given color, and the number of pixels that your output file contains (e.g. 1920x1080 in the case of HD). If the number of photocells in a particular channel matches the number of pixels of your output image you can say that you are not 'chroma subsampling' in that particular channel relative to the output raster size. That's why you can get HD 4:4:4 signal out of a Bayer sensor even if this type of sensor is 'chroma subsampling' in terms of what we mentioned in point 1. If you want to get 4:4:4 out of a Bayer CFA sensor you will need to oversample twice as many horizontal and vertical pixels as the desired output resolution. For example, you will need a sensor with 4Kx2K photosites for a true 4:4:4 2Kx1K HD output.

3. Finally, there is a sense of the term 'chroma subsampling' that refers to the signal coming out of the camera. The signal down the cable may, or may not have an even number of RGB data streams. If the number is even, then there is no 'chroma subsampling' of the signal. The notation 4:4:4 will mean then that the signal coming out of the cable dedicates equal bandwidth for each channel, RGB, or in the case of the YCC signal, that the same bandwith is dedicated to luma than to the two chroma channels. I think this is the sense that camera companies typically refer to when they use the term 'chroma subsampling' and notations like "444", "422", etc.

And it is important to keep in mind that this three senses of the term 'chroma subsampling' are, in principle, independent of each other. You can have 'no' chroma subsampling with respect to one of the senses, but not necessarily with respect to the others.
 
You must log in or register to reply here.
Back
Top