Personal View site logo
Truthfulness of short GOP superiority claims
  • 100 Replies sorted by
  • LPowellLPowell
    Sorry, but I doubt this particular test scene will show any significant differences between the three GOP-sizes. In order for AVCHD motion artifacts to become clearly visible, the shot must be carefully orchestrated:

    1. The scene should be brightly illuminated with high-contrast lighting.
    2. The background should contain fine details in sharp focus, requiring a high bitrate to encode.
    3. Foreground objects should have high-contrast edges and detailed textures.
    4. The camera should be panned slowly to generate a large amount of AVCHD motion vectors.
    5. The shutter speed should be 1/125 second or higher to minimize motion blur.

    Even when these conditions are met, it may be difficult to discern AVCHD motion artifacts without magnification. With a high enough bitrate, a long-GOP encoder will be able to track motion nearly as well as a short-GOP encoder. The short-GOP encoder will show its advantages in situations where the bitrate is maxed out.
  • GrunfGrunf
    According to my highly non-scientific tests, short GOP actually makes things worse when codecs pushed hard. Even more so when using only B-frames.

    Allow me to explain:

    This is uncompressed frame grab from 720p50 flick:
    http://www.luminasweden.com/vlcsnap-2011-07-24-22h44m57s135.png

    I used long GOP and B-frames. Bitrate topped out around 44Mbit. Even with this high bitrate and most efficient codec mode (B-frames, long GOP, high bitrate), the codec was on the edge due to wind shaking the leaves in random manner. But I pixel-peeped and couldn't see any macro-blocking anywhere. Only indication of maxed codec was slight "halo" around edges of leaves on sky when you turned up the contrast to relieve it. This sounds about right, as quantisation tables are still stock and high bitrate seems to go into encoding more accurate motion vectors instead.I couldn't see any signs of alleged "TV look" or 15-frame pulsing that short-GOP lovers claim.

    By using long GOP and B-frames, codec could allocate lot's of data to I-frames and still keep it within reasonable bitrate. If short GOP was used, codec efficiency would drop and it would be forced to throw data into encoding I-frames every now and then. All this data wouldn't fit into "reasonable" data envelope and higher quantisation values would be used and details would be lost.

    I believe that intra-frame codecs are superior quality-wise, but only if allowed to allocate A LOT more data due to horrible efficiency...like 4x or so. Crippling AVCHD to make "almost intraframe" codec at roughly same bitrate will IMHO just make things worse.

    I believe that key to good IQ from AVCHD is utilizing the codec where it's most efficient (stock GOP, B-frames) and tweaking the quantisation tables and bitrate as high as they will go.

    Regards,




  • onepointsixtwoonepointsixtwo
    @dkitsov

    How can you say this:
    "Sure, the only reason you say that locked off shots do not exhibit the phenomena is because when I will run the hand-held test of the same scene you will say that one cannot compare the clips because of the variable of motion of the camera operator.
    Still, as of now I accept your surrender, I guess you are too proud to admit that you cannot see a difference."

    And then say this:
    "By the way I am neither pro or against short GOP. I want to know the truth. I do not care about the opinions. I just wonder if short GOP is a Monster Cable of AVCHD on GH2."

    If you really did not care about the opinions then the concept of surrendering to your opinion would also not matter.
    You say you just want the truth but there is no truth. Bwhitz has stated that this is a totally subjective viewpoint yet there seems to be a concentrated effort to devalue that viewpoint. Surely the point of Ptool and other hacks for other so called consumer products is to allow users the freedom to use those products as they see fit. In this instance, to produce moving images that we, as individuals, prefer the look of. We are all searching for solutions that fit within are own set perceptions. Just enjoy that fact and go shoot some great imagery.
  • madactmadact
    I picked 6, 3, 12 straight off the bat.

    Seems, looking at comments that I'm not the first to think so... four concurrences (plus one picking B=3) and two totally different so far... I'll be interested in the 'reveal', but there's definitely a visual difference.

    FWIW I'm neutral on GOP for GOP's sake... Could be B is long-GOP and had more bandwidth or short-GOP and less motion estimation, all I know is that the other two look very 'swishy' and blurred by comparison.
  • bdegaziobdegazio
    I agree with lpowell that these clips are not very effective at testing motion capture.

    The problem that I saw as soon as I started using the GH2 was that in a high-detail, high-motion shot the motion in darker areas of the shot can be seen to stutter at a rate related to the GOP (Group Of Pictures) rate. The problem was so severe I almost returned the camera.
    Here are a couple of clips that I've cut together from my own own shooting with the original and hacked GH2 which I think shows the problem clearly. There is a lot more detail on the vimeo page:




  • fehafeha
    @bdegazio
    great, did you apply any GOP ?
  • bdegaziobdegazio
    My notes on the vimeo page describe the situation, but basically I think it's GOP6 or GOP3 for the 2nd shot. If anyone would care to check, the original file is also on vimeo (
    ) and can be downloaded. Unfortunately I can't run StreamParser myself.
  • cbrandincbrandin
    There is a possibility that short GOP's may improve dark details. The GH2 codec uses a minimum quantization value for P and B frames of 20, and a minimum quantization value in I frames of 18. Lower quantization values usually mean better shadow detail. Short GOP's will mean that lower quantization values are used more often. This would show up particularly in dark areas in motion where highlights are encroaching on previously dark areas; and probably even more so the other way around. It could also show up in fast moving details, like moving water.

    Chris
  • stefanosstefanos
    thank you Chris Brandin. This is just the kind info that is needed for us to better understand how GOP and quanitization works! I am really looking forward to advances in quantization since your results show big potential for improvement in macro-blocking too.
  • dkitsovdkitsov
    @onepointsixtwo I thought it was a universally recognized inter-webs indicator of sarcasm when the sentence starts with "Sure"
  • dkitsovdkitsov
    @bdegazio One of the reasons I was anxiously waiting for the @Vitaliy_Kiselev hack of GH2 was that very same problem you have described - pulsation in the shadows corresponding to the length of the GOP (well back then I did not know of the GOP length anyway) roughly 2 times a second in a 24H mode.
    This problem is now largely non-existent with the high bit rate regardless of GOP length.
  • dkitsovdkitsov
    @cbrandin Thank you Chris for the non-opinion but something factual.
  • cbrandincbrandin
    bdegazio,

    I curious about something in your test film. In the earlier part, is the water moving in 1/2 second pulses? And, in the latter part, is it smooth (i.e. moving in 1/24th increments) or just better (i.e. moving in 1/8 increments if the GOP is 3). I'm trying to figure out if the improvement is in all frames, or just because of the higher percentage of I frames.

    By the way, I've noticed that many responses start with "@name". Can somebody tell me how to do that?

    Chris
  • telkitachkitelkitachki
    @+cbrandin = @cbrandin
  • Mark_the_HarpMark_the_Harp
    Chris, you just type @(whoever) and it turns into a link for that person
  • Vitaliy_KiselevVitaliy_Kiselev
    >By the way, I've noticed that many responses start with "@name". Can somebody tell me how to do that?

    Just type this like @cbrandin .
    After this you can set in your preferences to inform you by email about each mention of your nickname :-)
    I removed on screen popups, but hope to get them back after script update.
  • bdegaziobdegazio
    @cbrandin

    " In the earlier part, is the water moving in 1/2 second pulses?"

    Yes. The GH2 720p60 captured the motion in the reeds quite well, but gave up on the water, maybe because it's darker, or lower down the frame.

    To clarify, however, these are two completely separate shots, taken about a week apart. They happen to be similar in appearance, sans the pulsing. The first shot is pre-hack, the 2nd comes from experiments with high-bitrate, low-GOP (I think perhaps using kae's 3GOP settings.) I can't run StreamParser on my Mac under Wine, so I can't confirm the GOP though, only that it plays back at 33mbps.

    "And, in the latter part, is it smooth (i.e. moving in 1/24th increments) or just better (i.e. moving in 1/8 increments if the GOP is 3)"

    I'm not sure, but it's definitely NOT pulsing at 1/2 second period. And to my rather practiced eye (I spent 20 years eyeballing shot sync for IMAX, and I now teach in an animation school where I'm always judging frame rates and smoothness of motion) it looks like a solid 24 fps. My theory about it (you would know a lot better) is that the higher bitrate and lower GOP means that the B frames are less 'stressed" and therefore more accurate.

    "By the way, I've noticed that many responses start with "@name". Can somebody tell me how to do that?
    "
    I'd like to know myself - I've just been cutting and pasting!
  • cbrandincbrandin
    Gee, that was simple - I thought there was a hidden "reply" button somewhere:)

    @bdegazio (my first use of newly acquired knowledge)

    I think it's because the water was low in the frame. I have noticed that the codec seems to figure out when it is running out of bandwidth low in a frame and just abandons additional processing. I suppose that is why some codecs have a 2-pass feature.

    Chris
  • Vitaliy_KiselevVitaliy_Kiselev
    >Gee, that was simple - I thought there was a hidden "reply" button somewhere:)

    I want to add quote button. But do not like present code.
    Also thinking about move to Markdown.
  • kaekae
    @dkitsov

    I think to truly answer your query anyone would have to see the original clips without your transcode to see a difference. I've been testing various GOP lengths for months first with the Gh1 then the Gh2 using 3, 4, 5, 6 and 12. With motion I ALWAYS notice a subtle difference and it SEEMS the lower the GOP the more filmic it SEEMS to me. No one here (I think) has ever said short GOP is superior. Some of us just PREFER it over long GOP - me very much so. So the answer to your transcoded - non-original .mts file test is: I don't have a clue. And your footage holds no interest for me whatsoever. It's what I like in my footage that matters. If you don't like short GOP why waste a second of your time on it? Just load 12 or 13 GOP settings and get to work. This whole debate is sophomoric. It should have never been a debate, it should have been an 'option'. Others, mainly detractors, have made it a cause célèbre. I thought all of us were busting our ass to test settings so that we could offer options to artists not theorums to theoreticians?
  • Aleksandr_OleynikAleksandr_Oleynik
    +1
  • dkitsovdkitsov
    @kae
    Of course you would want to see original MTS. This way you would be able to run it through a parser and say "of course this clip is better than that, as it has lower GOP" The best I can do for you is to give you is an i-frame cineform if you are not satisfied with h264 available for download on vimeo.
    However sophomoric this debate might be, it does serve a purpose of trying to find out if the 3GOP really is better and if it is worth of spending resources of the codec/card to encode all those extra i-frames or can we spend those resources on the improving overall image quality.
    To repeat myself yet again I do not see a difference and I am curious how many of the populus of a self-selected group are able to do so, in case something is wrong with my eyes. This is more of a statistical study then a technical.
    If anything as Chris has hinted there might be benefit from a technical point of view of using more i-frames allowed by a shorter GOP because of a lower quantization.
    Ah... but then again if one allows oneself to be reasonable how can one then be a martyr and carry a cross for all of us simple-minded folks, who do not posses the magical eyes that are able to detect such subtle differences.
  • stefanosstefanos
    @dkitsov
    I understand where you're coming from and I think it's great that you're testing this objectively as I'm very interested, but really there's no point making it your goal to disprove people and disprove subjectivity. For all you know, it might be a placebo effect, they might have convinced themselves that they're seeing a difference, but who cares, placebos are great! " ...and say"ofcourse this clip is better than that, as it has lower GOP" to kae's defense, he has made it clear it's not about superiority or being "better". Then you say it "serves the purpose of...finding if GOP3 is better and if it is worth spending resources" Again, GOP3 fans are not debating being better so there's no real argument here, and as pointless as you may find it, kae and other people are free to spend their own time and own resources wherever they want.
    @GOP3 fans, the mistake that you guys make is the following: you should refrain from comparing GOP3 with film and saying that it's similar to ALEXA, FILM, RED. You shouldn't say that these settings are more for cinematographers and others are more for videographers. I mean com' on, everyone including videographers aspires to come closer to the qualities of ALEXA,FILM etc and by saying that kind of stuff, you make people who think high GOP is better, feel guilty. And without proof, direct tests and comparisons with FILM/ALEXA etc, such claims are BS.
    we are all mature here and objectivity and subjectivity can coexist.
  • ButtButt
    @stefanos
    lowGOP-Comparisons with the RED and Alexa, etc. are subjective and therefore not dangerous for other
    one takes what one likes most: lowGOP or longGOP
  • madactmadact
    So, is anyone else actually going to take a shot at the original question?
Add a Comment