Questions About Professional and Prosumer Codecs

Discussion in 'Digital Video' started by Chris7, Apr 5, 2009.

  1. Chris7 macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #1
    I Have a couple questions about pro/prosumer codecs…

    1. Panasonic’s DVCPRO HD uses 4:2:2 chroma subsampling. From Wikipedia, I see how 4:2:2 subsampling is superior to the 4:2:0 of HDV or AVCHD. (And it’s obvious why DVCPRO HD’s intraframe compression is superior to the interframe compression of AVCHD or HDV.) But both appear to be 8-bit formats. What exactly is the difference between 8 and 10-bit codecs, if it does not appear to have to do with the color subsampling?

    2. Why don’t more cameras use high-quality codecs? Sandisk is now making solid state hard drives for laptops that can record 200 MB/sec. One that stores 256 GB is about $500. So the technology is there, and its cheap. It would seem that something like AVC-Intra 100, for example, could be cheaply implemented as a prosumer standard. But the cheapest camera using a codec of this quality is $10K and records on expensive P2 cards ($2500 for a single 64 GB card). I don't know of another high quality codec that is any cheaper. Why do you suppose this is? Editing software support? Marketing tactics (make people buy more expensive cameras in the line)?

    Thanks in advance,
    Chris
     
  2. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #2
    Chroma subsampling is a ratio between the amount of luma and chroma info recorded in a format. Color bit depth (8-bit, 10-bit, etc.,) is how many different colors can be reproduced or, to think of it a different way, how many shades of gray there are between pure white and pure black. 8-bit ranges from 0-255, but in TV land 16=black and 235=white. 10-bit goes from 0-1023 which, obviously is a much larger range resulting in a cleaner picture (less chance of color 'banding' for example). For 10-bit in TV land 64=black and 940=white. Values lower than TV land black and above TV land white are called "super-black" and "super-white" respectively. Many cameras will record in these "illegal" areas so that in post you have more picture information to work with and during color correction you can "reign in" everything to be broadcast safe.

    The reason we don't see more high quality codecs is, basically, because they are expensive to create and implement. Having all that circuitry on board a camera takes up space and can create a fair amount of heat. The early Red Ones, for example, had over heating problems as well as very load fans (two things you don't want in a camera). As time goes buy we'll see those codecs "trickle down" just like all tech does. What is expensive now will be almost common in a few years. There are also marketing/business considerations as well as these companies, like all companies, can't stay in business if they cannibalize their own products. The companies also have to think long term and can't create "throw away" devices because professionals buy their gear and expect to get years of service out of it. Proven and reliable gear, as well as workflows, is the corner stone of this industry. To this day there are still PowerMac G4 towers running OS 10.2.x and equally as old versions of Avid pumping out content for TV. Why? Because the workflow is built proof and the gear is long since paid for. ;)


    Lethal
     
  3. KeithPratt macrumors 6502a

    Joined:
    Mar 6, 2007
    #3
    Incidentally, DVCProHD uses luma subsampling, too — and heavier than HDV in 59.94Hz.

    P2 cards were used as Panasonic could vouch for their reliability. To professionals, the added cost is less important than the assurance of stability; thus their use only in the professional line.

    To add to (/repeat) what Lethal said regarding AVC-Intra: your question is not one of how the videography industry specifically works, but of the whole of commerce. Companies need to recoup investment costs and they need a prestige line. They accomplish both by pricing their newest and best technologies high. With the prestige prestige, prices generally remain high through the product's lifespan, where in the garden prestige prices generally fall as a better product is introduced.

    $10K is a lot less than the previous cheapest AVC-Intra camera. And you say "prosumer standard" as if that's an end to it. Industry keeps moving (hence the name) and everyone will want whatever inevitably surpasses AVC-Intra as soon as it's announced.
     
  4. spinnerlys Guest

    spinnerlys

    Joined:
    Sep 7, 2008
    Location:
    forlod bygningen
    #4

    Tell me about it. I thought I was one of the only one left to have to work with them. I mean it's a mean machine compared to a MacPro and FCP (which can't just show a mask on top of a video without rendering, like bars).
    It just gets complicated when you have half a dozen of external drives and Disk Utility can see them, but no way of mounting them.
    Or Multi-Cam editing. Slow, slow, slow.

    But sorry Chris7 to hijack your thread like this, I was just so surprised by Lethal's comment, that I had to say something.


    To one of your questions: maybe P2 cards are more durable for in the field camera work, and the drive can be much smaller than something like an SSD socket or something.
     
  5. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #5
    I've been reading more this afternoon to try to understand chroma subsampling better. Perhaps it's just a complicated thing. I keep hearing the Y value refereed to as luma (which makes sense), but then the Cr and Cb referred to as "color difference." I keep on wondering if the Cr somehow refers to red and the Cb to blue, but then, where would the green be?...
    Does a RGB colorspace take up more data than a Y'CrCb 4:4:4?

    Anyway, even 4:2:2 seems like the loss of a lot of information (a given color is averaged across two pixels?). Does ProRes use something like this?
    I think I understand you here. I think the bit thing will make more sense as I come to better understand color subsampling.

    OK. So the reason prosumers don't get to use these codecs is due to both technological limitations and marketing factors. Good to know it's not marketing only.

    Thanks for your time,
    Chris
     
  6. bigbossbmb macrumors 68000

    bigbossbmb

    Joined:
    Jul 1, 2004
    Location:
    Pasadena/Hollywood
    #6
    Red+Blue+Green=Luminance

    If you know Red, Blue, and Lum you can figure out Green.

    Although 4:2:2 seems like you are throwing out a lot of info, our eyes perceive luminance a lot more than color information. Visually it looks almost the same. The reason we use sub-sampled chroma is to get by with lesser hardware. 4:4:4 video takes a ton of storage and bandwidth. FCP will not work with 4:4:4 unless you have a professional I/O card and the corresponding codecs.

    And yes, ProRes is 4:2:2.
     
  7. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #7
    Interesting. I thought it was 4:2:2, but I have a limited understanding of chroma subsampling. Maybe a different kind of 4:2:2 than the better 4:2:2 codecs?

    Makes sense to me, though of course I wish the pro codecs were cheap now. I read that the price of solid state hard drives has come down incredibly fast over the last few years. The price of P2 cards still seems a little high to me considering the price of SS HD’s have dropped so much (e.g. the Sandisk laptop SS HD’s).


    I was missing this. Started to try to figure out the RGB to Y’CrCb equations on Wik but didn’t get far. Many thanks.
     
  8. mBox macrumors 68020

    Joined:
    Jun 26, 2002
    #8
    As a professional Motion Designer/Animation/Digital Multimedia Specialist, that was the most concise answer Ive ever seen written on the internet :) Thanks after all these years working in the background of broadcast your words put a few things puzzling me that I was too lazy to even deal with. Of course I never really had to deal with it being just a creative :(
     
  9. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #9
    mBox,
    Glad what I said was clear. It's easy to start trying to explain this mumbo-jumbo in too much detail and watch people's eyes glaze over. Sometimes I have to remind myself that not everyone finds this "nuts & bolts" type stuff as interesting as I do.

    I think what KeithPratt was saying about DVCProHD is that codec is not full raster. So, for example, even though DVCProHD 1080i60 displays at 1920x1080 the actual resolution is only 1280x1080 (and if you are shooting from w/the HVX the CCDs are only 960x540 but pixel shifted to make more efficient use of the sensors). HDV 1080i60, by comparison, is 1440x1080 so, all other things being equal, HDV 1080i60 has more luma info than DVCProHD 1080i60.

    Y'=Luma (not luminance. There is a difference which is explained in complete geekdom here, YUV and luminance considered harmful
    Cb= Blue - Luma
    Cr= Red - Luma

    I think R'G'B' and Y'CbCr @ 4:4:4 would take up the same amount of space because no subsampling of the Y'CbCr has taken place, but don't quote me on that.

    For more info check out these links
    Less Geeky:
    Understanding Color Sampling from DVXUser.com.

    Geeky:
    Chroma Sampling: An Investigation by Graeme Nattress.

    Very Geeky:
    Chroma Subsampling Notation by Charles Poynton. Check out the rest of his site as well.

    Bonus link:
    What is Y'CBCR? at mir.com.


    Lethal
     
  10. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #10
    Makes sense. Thanks for the clarification.

    I appreciate the links. I get really into the "why" of things, and this one was really puzzling. I've not read them yet, but the one by Graeme Nattress uses the word "bit" 48 times. Should be enough to satisfy my curiosity regarding how chroma subsampling and bid depth relate...

    -Chris
     
  11. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #11
    I read the above links. Just a few more questions, if anyone's interested...

    1. My understanding of RGB is that it can desplay colors that Y'CbCr cannot -- saturated yellows and blues, for example. Is this completely true, or would it be more accurate to say that Y'CbCr can create these colors, but a CRT TV just cannot play them, so these colors are not "legal?" I guess one way I could better understand this is to ask, "Can one transfer RGB to Y'CbCr 4:4:4 and back to RGB without losing information?"

    2. I have read that the human eye is much more sensitive to luma than color, and that it is also more sensitive to green than red and blue. In a 4:2:2 chroma subsampling, the luma is obviously represented twice as much as chroma (the luma being represented pixel for pixel and the chroma averaged horizontally over two pixels). But is the green also represented more than the blue and red, or are all the colors represented the same?

    3. Does a 10 bit Y'CbCr codec give a range of 64 to 940 levels of intensity of each of red, green, and blue (by "levels of intensity, I mean like from no green to totally green)?

    Thanks,
    Chris
     
  12. CaptainChunk macrumors 68020

    CaptainChunk

    Joined:
    Apr 16, 2008
    Location:
    Phoenix, AZ
    #12
    P2 cards are built for rock-solid reliability. Never once have I encountered a bad P2 card. And I suppose one would expect durability when the cards are so expensive. They have gotten a bit cheaper over the years, however. I remember when a 4GB card cost close to $1,000 when the first P2 cameras started shipping.

    I'd like to see cheaper P2 cards too, but since Panasonic continues to keep them proprietary, they charge whatever they want and they know that their pro customers will pay it. There has been talk about third-party P2 cards (like Hoodman and Fujifilm), but I don't think an actual product has surfaced from that yet, at least in the US.
     
  13. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #13
    I'm gonna preface this by saying this is beyond my typical knowledge base so I reserve the right to be totally wrong. :D

    1. Y'CbCr can contain values that are 'illegal' for use w/TVs. Either because the TV can't display them or because the illegal values can cause a disruption in the transmission/display of a broadcast signal. For example, if the luma is too high it can interfere w/the audio portion of the TV signal. As for going between R'G'B' and Y'CbCr 4:4:4, I'm not sure, but I think it there is a bit of information lost since R'G'B' goes from 0-255 while a Y'CbCr video signal is equal to 16-235 (in R'G'B' terms). An interesting thing to note is that you can have colors in Y'CbCr that don't exist in R'G'B' because the Super Black and Super White areas of Y'CbCr don't have an corresponding R'G'B' value. For example, if 16 in Y'CbCr is equal to 0 in R'G'B' then 0 in Y'CbCr doesn't have an R'G'B' equal because R'G'B' can't go below 0.

    2. Green is represented the most. Images used for cameras are also more sensitive to green and least sensitive to blue. This is why typically the blue channel will have the most noise in it because the camera will have to electronically add gain to the blue channel to keep it properly balanced w/red and green.

    3. 10-bit gives you a range from 0-1023 but only the values between 64-940 are legal for video work.

    W/regards to P2 cards, something to keep in mind is that each P2 card is actually made up of 4 solid state memory cards running in RAID 0. At the time this was the only reliable way to get fast enough write times to flash memory, but obviously cards are getting bigger and faster so Panasonic might have painted themselves into a corner if they don't figure out a way to add better value to their P2 cards.


    Lethal
     
  14. bigbossbmb macrumors 68000

    bigbossbmb

    Joined:
    Jul 1, 2004
    Location:
    Pasadena/Hollywood
    #14
    The controller chip does more than just stripe to 4 memory cards. They actually perform write verification and correction on the fly. If the controller makes a mistake on the write, it'll be caught during verification and corrected without dropping a frame.
     
  15. KeithPratt macrumors 6502a

    Joined:
    Mar 6, 2007
    #15
    Neither RGB nor YCbCr are really colour spaces themselves. It's sort of like saying "it's a car". Rec.709 is a colour space. It's the one used for HDTV. Saying Rec.709 is like saying "it's a 2006 Ford Focus ST — the one with the go-faster stripe."

    Other common colour spaces are Rec.601 (SDTV) and sRGB (jpeg photos). Each of these can be either RGB or YCbCr (yes, lots of jpegs are actually YCbCr). And there's Adobe '98 or DCI-P3 or some other unspecified "wide gamut". As far as I know, there's nothing stopping these from being YCbCr — they just tend not to be.

    If you are thinking in terms of digital cinema cameras, like the Sony F35, Panavision Genesis, and Red ONE, they can shoot Rec.709 or a wider gamut. For cinema it'd make sense to shoot wide gamut. The same could be true for TV, since the the wider gamut can be useful in grading and then squeezed to Rec.709 for broadcast; but the post- workflow might be better suited to using Rec.709 start to finish. Horses for courses.

    Oh, and CRTs display in RGB but likely would not have a wide enough gamut to display even Rec.709 in its entirety.

    You're welcome.
     
  16. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #16
    I think one thing that threw me off regarding 4:2:2 chroma subsampling, is the macprovideo.com tutorial on color correction in FCP6. Jamie McCallister basically says that only red and blue are subsampled, and the both the luma and the green are displayed fully in the 4 of 4:2:2. He’s supposedly an expert colorist and perhaps he was just trying to simplify things, but this is inconsistent with everything else I have read, and also doesn’t make much sense. I tried going through the RGB to Y’CbCr conversion equation to see if green is somehow represented more in the Y than in the Cb or Cr, but I really can’t make heads or tails out of that equation.
    On a lynda.com tutorial, Larry Jordan shows a FCP vectorscope with extra lines drawn around the red, magenta, blue, cyan, green, yellow boxes. He’s saying that the area around these lines cannot be represented in a Y’CbCr colorspace, and gives an example when directing live TV of pointing a TV camera at someone wearing a “heavily saturated blue magenta” dress and her dress “would go either solid blue or solid red.” I’m not sure, but I think he’s implying that, in RGB, anything in the vectorscope could be represented. On another tutorial, he points differences in the way an computer monitor and a video monitor display color – different gamma and white levels, for example. I was thinking he was just simplifying things so he didn’t have to talk about rec. 601 and such, but he also says that video displays Y’CbCr. Pretty confused on this one, now that you say a CRT displays in RGB.
    Thanks again.:)

    Sandisk’s 256MB solid state laptop drive will write at 140MB/sec for $500 retail. It would seem 256MB is enough to write about five hours of AVC-Intra, or about two and a half hours of “4K” Redcode 28. I don’t this would ever really happen, but if the camera company did not want to license a codec, they could allow the user to select between a few, like ProRes HQ, DNxHD, or CineForm (Film Scan). A 256MB drive would record somewhere around two and a half hours of ProRes, DNxHD, or CineForm Film Scan (at 4:2:2 10-bit 1080x1920). You could empty the drive in under an hour with a FW800 port.

    Toshiba is soon to release the first 512MB solid state laptop drive, and it is said to write at 200MB/sec (don’t know the price). The price of flash media is moving down fast. Here’s an excerpt from an interview with John Galt (who worked on the Panavision Genesis) talking about pixels and digital film…

    http://magazine.creativecow.net/article/the-truth-about-2k-4k-the-future-of-pixels
    We just introduced the SSR-2, a dockable solid state recorder. We can record up to 84 minutes of uncompressed 1920x1080 at 4:2:2 or 42 minutes at 4:4:4 That requires almost three quarters of a terabyte of solid state flash memory. (We didn't consider hard drives because they just aren't reliable enough.)… When we started developing it three years ago, the flash memory cost alone to give you that recording time would have been $68,000! Of course, what happened during the few years of development is that the price of flash dropped to 1/10 of what it was when we started…
     
  17. P-Worm macrumors 68020

    P-Worm

    Joined:
    Jul 16, 2002
    Location:
    Salt Lake City, UT
    #17
    I think this picture helps to illustrate the difference between things like 8-bit and 10-bit. With a higher bit depth, you have much more room for shades so you don't get the banding in the first image. The middle image shows dithering which just blends the colors a bit to make it appear more like a higher bit depth.

    [​IMG]

    P-Worm
     
  18. KeithPratt macrumors 6502a

    Joined:
    Mar 6, 2007
    #18
    Green's presence is a result of being able to work it out from the luma (Y'), blue (Cb) and red (Cr).

    This Larry chap may be saying YCbCr and meaning Rec.601/709, or he may be meaning out-of-gamut colors that are a result of the concept of subsampling. But the latter are more like anomalies, they don't define an entire gamut/colour space as far as I know.

    CRTs, LCDs and plasmas all display in RGB. (If you get really close you can see the individual little red, green and blue lights that make up each pixel.) YCbCr is just a method of storage and transportation, it's not an image as we know one before some maths is applied.
     
  19. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
  20. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #20
    Lethal's links gave me a pretty good understanding about chroma subsampling. The simple graphic on this one, in particular really helps, as it shows the averaging of the chroma across pixels, using a variety of colors: http://www.dvxuser.com/articles/colorspace/

    I would not be still having questions about the green if it were not for that Macprovideo.com tutorial with Jamie McCallister, where he's calling the Y the luma and green, and the Cb and Cr the blue and red. I know that this is inaccurate and that all three components carry red, green, and blue. But it does make me wonder if the Y somehow carries more green, so when the Cb and Cr are averaged across the two pixels, green somehow gets subsampled less in 4:2:2 (regardless of how much green was in the initial shot). But would make absolutely no sense, as it takes all three colors to make up a color. Maybe the Bayer sensor is totally irrilivent to color subsampling, but the Bayer sensor pattern makes me wonder if there’s any truth to what McCallister is saying (From wik: "Bryce Bayer's patent called the green photosensors luminance-sensitive elements and the red and blue ones chrominance-sensitive elements. He used twice as many green elements as red or blue to mimic the human eye's greater resolving power with green light.")

    Besides the McCallister tutorial, everything I've read about chroma subsampling goes something like this: If in RGB, red, green and blue are represented equally, than in 4:4:4 Y’CbCr, red, green, and blue are all still represented equally. In 4:2:2 Y'CbCr, green, blue, and red are still sampled at the same rate – even though they are all subsampled, one color is not subsampled more than another (even if the green sensor of a 3CCD camera typically gets more info). The chroma is simply averaged over two horizontal pixels, whereas the luma gets its own pixel. And that all there is to it. Do I have this right?

    BTW, I’m not knocking Larry Jordan for his oversimplification. His FCS tutorials on lynda.com have been incredibly helpful to me, and are the most thorough, organized, and pertinent I’ve seen.
    Thanks for the graphic.

    -Chris
     
  21. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #21
    That article is good about breaking down it down, but keep in mind the graphs are solely related to subsampling and do not accurately reflect how Y'CbCr looks.

    All components don't carry red, green, or blue though. Y' is luma and Cb and Cr are difference channels. Cb is Y'-Blue and Cr is Y'-Red. Red, green, and blue don't exist as discrete channels in Y'CbCr.
    Google has a selection of this book, Color Correction for Digital Video, and if you go to page 24 it implies that the green channel is encoded in the luma channel. When reading about Y'CbCr you might have see the term "weighted sum" and that means that not everything is equally counted. While I don't know if the information contained in the Green channel is uncompressed, there is definitely more info from the Green channel than the Red or the Blue.

    Subsampling doesn't change the color balance it just changes how much resolution is in a given color channel. For example, when you look at some of the pictures in the Graeme Nattress link the images are separated into Y', Cb, Cr and there is obviously much more picture info (i.e. less noise) in some of the channels than others but when combine all the channels the picture doesn't look out of balance.

    Correct. The camera sensor, the recording codec and the display device are all separate pieces of the same chain.

    Larry Jordan hasn't always been very reliable when discussing the nitty gritty tech details, in my experience.


    Lethal
     
  22. Chris7 thread starter macrumors 6502

    Joined:
    Aug 8, 2008
    Location:
    Lost in Thought
    #22
    I realize I said this wrong above. As I understand it, the Y' is basically shades of grey, the other two channels carry the color.
    OK. THIS is the part I've been most curious about, and the part I'm getting hung up on. Can you explain this in detail (if you're still interested)?

    Also, looks like you've read a bit. Are there any books you'd recommend about how to shoot and edit? You mentioned in another thread that "The DV Rebel's Guide" was good. I'm also interested how to make advertizements.

    Thanks again,
    -Chris
     
  23. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #23
    I don't have very many production books (the DV Rebel's Guide is actually my only one). Most of the stuff I have is for post. For editing anything by, or about, Walter Murch is insightful and for more technical and color correction type stuff check out any of Steve Hullfish's books.


    Lethal

    EDIT:
    What exactly happens to the green (if it's sub sampled or not) I'm not exactly sure. But if I figure it out I'll share. ;)
     
  24. LethalWolfe macrumors G3

    LethalWolfe

    Joined:
    Jan 11, 2002
    Location:
    Los Angeles
    #24
    P2 E-series cards will be much cheaper retroactively named A-series cards and have a shorter rated life (5 years of full capacity, daily use). While is this good news the cards still aren't inexpensive.
    While the P2 cards are more complex and robust than regular flash media, which makes them more reliable for repeated use, I think the industry is headed away from reusing memory cards and heading towards writing once and putting it on the shelf. Once flash memory prices start to become in line w/tape prices I think we'll see people shooting to a flash card once then shelving it or handing it off to a client just like they do tape.


    Lethal
     
  25. bigbossbmb macrumors 68000

    bigbossbmb

    Joined:
    Jul 1, 2004
    Location:
    Pasadena/Hollywood
    #25
    That is definitely the direction we're going in...and we're really not that far off.

    especially with the EX1/EX3 ability to shoot to SD cards with an expresscard adapter. It makes you 16GB of media only $25-$30. These should come down to professional tape prices fairly soon.

    There's also the option of shooting P2 and having the media wrangler copy the disk image to flash media instead of simply copying the folder a RAID1.
     

Share This Page