Personal View site logo
Official Low GOP topic
  • 1003 Replies sorted by
  • @kae
    i noticed that the last 4 frames are always (even the ones with normal footage) not ingested by 5dtorgb (it causes the freeze of transcoding) and most players have many difficulties to play the clips (for me it's impossible to get to the end). No issues or problems after transcoding.
  • @APXmusic
    thanks for Tipp - Media Player Classic runs - VLC does not
  • @robbin You need to register/login to get them... Al
  • @Kae

    My files were all real-world footage of the "lawn of death" where I found the gh1 native 24p problems. :)
  • kae@ with what you look the clips to see of are fluid or stuttering occur ?
    with the 30p 60i not problem on iMac 27 at 60hz but 24p 50i footage need switchResX to bring the monitor to 50hz.
    Now i test on a projector in 1080p cinema 24fps.
  • @svart
    "My files were all real-world footage of the "lawn of death"

    Got it. I only have the problem at the end of the clip with death chart footage too. Try high detailed footage that you would normally shoot, not typically codec buster stuff and see if you get the glitch at the end.

    @Ulisse
    I use Windows Media Player in WIndows 7 64bit. Then I ingest at DNXHD-175 into Avid Media Composer 5.5 But I'm not seeing any stutter on playback, Svart is seeing that.
  • Just ran Kae's 3 GOP setb.ini (22/07/2011) tes on an hour long recording. The first file in Streamparser shows up strange result. Look!
    gh2test 24p kae - hour record test - 1st file.png
    1320 x 709 - 77K
    gh2test 24p kae - hour record test - 2nd file.png
    1346 x 701 - 88K
  • @driftwood

    have you the setb in the PTool3.61D loaded with "B" or did you enter by hand?
  • I think this is a bug in Streamparser. Maybe I have an overflow happening because of the high bitrate. Could you please post a picture of the time-based analysis?

    Chris
  • @Butt
    Loaded with B
  • Driftwood, like Chris just mentioned, switch to the time mode setting and screen capture this.
    This is not a camera issue, but you sure found a bug with streamparser.
  • @driftwood
    this?
    setb.zip
    439B
  • @ driftwood

    If you look at the video bitrate reported it's a negative number - that usually happens because an integer variable overflowed. I suspect I need to change a variable from int32 to int64 somewhere. I wanted to see the timing analysis just to make sure that's all it is.

    Chris
  • @cbrandin
    Time based attachment of first file of 5 files (4 gig FATDOS limit breaks files into chunks as you know!) Looks alright it seems. BTW: Kae - picture qaulity on moving subjects seems excellent.

    @Butt
    Yes.
    gh2test 24p kae - hour record test - 1st file - TIME BASE.png
    1291 x 664 - 93K
  • Testers only:

    Been testing different 1080 24P 3GOP combinations for the last five hours - have tried adjusting quantizer encoding settings, encoder frame limit, frame buffer sizes for 24P, and 24P video buffer settings and nothing has given me a more stable result than my previous settings. I did manage to up the 24H video bitrate to 65Mill (max for ptools 3.61d) and it seems to be as reliable as 60mill. The advantage is a higher data rate encoding. I'm getting 64mb/s on the death chart and normal footage is encoding at a higher rate. Can I see a major visual difference? Not sure. It's all so clean at this data rate, it's hard to tell short of shooting rez charts, etc. I've attached the 65M 3 GOP settings below if someone wants to test. You WILL get 10 frames of blockies at the front, I am not getting Svart's end glitch on playback and have no problem ingesting the full clip into the Avid, but others might, so be forewarned - ROLL EARLY and CUT LATE with these settings. And please shoot the NORMAL footage you will get paid for rather than just lawn, trees and the death chart to see how you like the look.

    I think I see a whisker more aliasing at 65M but this may be due to the encoder being able to salvage more detail. I've been shooting with sharpness at 0.

    Still hoping others can eliminate the 10 frame glitch, I couldn't eliminate it with all the combos I used in the last 5 hours so I'm still stumped on that one. FWIW: I'm using a Dane-Elec 32G class 10 card.

    EDIT: SEE MODIFIED SETTINGS PAGE 6 FOR MORE STABILITY IN OTHER 720P/1080I MODES!
    KAE1080_24P_65M_3GOP.zip
    505B
  • @driftwood
    use this
    and you have no problem
    and - incam-play transfer works
    setd.zip
    487B
  • @cbrandin
    Defo integer prob as you can see the fifth and final file written (at around half a gig in size) reports okay.
    gh2test 24p kae - hour record test - 5th file - stream frames. check EOF write.png
    1297 x 665 - 90K
  • @Butt
    Thanks I'll give it a whirl on another hour test.
  • @butt
    "use this
    and you have no problem
    and - incam-play transfer works"

    How'd you set 24H buffer to 70mill? Ptools won't let me do that? Will try your settings now. Thanks.
  • @Butt

    I see what you did now. What caps the data rate is: "Video Bitrate 24H=60000000". These 3 settings: "Video Bitrate 24H ADD=70000000
    Overall Bitrate=70000000
    Overall Bitrate 2=54000000" have no effect and you can just uncheck them.

    Try the settings I just attached above. It will give you 65mb rather than the 60mb you're getting with yours.
  • @cbrandin
    Yep the same first file now reports fine in stream frames/bitrates. cheers!
    gh2test 24p kae - hour record test - 1st file - stream frames. check cbrandin int 64bit.png
    1300 x 678 - 91K
  • @kae
    Gonna test your upgraded 'Butt' test . Incidentally, how long was your last satisfied 65mb stability test for?
  • I respect much of the work on this topic.

    However I feel the bigger things are getting lost in all these distractions and I am still no the wiser over whether to adjust the GOP for better image quality or worse.

    GOP 3,6 or default for low light shooting, which is the best choice?
  • I haven't found these threads to be a distraction at all - they are why I began to look into quantization.

    Also I learned that there is a tradeoff - lower Q means better detail (especially in low light), but worse temporal behavior, and visa versa. There is clearly a balance that needs to be established. If it weren't for both the low GOP and MJPEG threads this would have never come to light for me.

    So, thanks guys!

    Chris
  • @EOSHD
    6 is closer to 3 in aesthetics, and closer to 12 in efficiency;
    A simple 42mbps modification is enough w/ 6 for densely detailed shots.
This topic is closed.
← All Discussions