Personal View site logo
Make sure to join PV Telegram channel! Perfect to keep up with community on your smartphone.
GH2 - 2.5fps AVCHD 24h Timebuster/TimeHBusteR 2.0 settings - The day is not over!
  • 323 Replies sorted by
  • yep, I agree @KeithLommel and @Hallvalla "sacrificing 1080i and HBR to create the timelapse mode, and retaining fully featured 720/60p and 1080/24p."

  • @duartix 'GOP Related' patch in ptools allows you to stack weights to P/B frames, try on whatever gop you settle on to leave it to ;-

    1, 0, 0, 1, 0, 0

    which equals all weight to i frame. See what happens. :-)

  • @duartix, as a PAL person I am happy to have a resilient HBR 25P and using 24P for the timelapse functionally. I would also like the best slow mo as an option patch with lesser 25P HBR performance.

  • How do I use these settings in GF2? It doesn't have 24p, but I could sacrifice 720p for this.

    • @LPowell said: "HBR is 1080i, there are no HBR-specific settings" Thanks for that info, but it's an almost, I believe 25p HBR still gets it's GOP from 24p.

    @driftwood : I tried that about 3-4 days ago but either it didn't seem to do anything at all, or I was making the same testing mistake that usually derails me (forgetting to test both PAL & NTSC). I tried default vs 1,1,1 vs 12,1,1 and they seemed to keep their sizes every time. I'll retry it for HBR.

    @kavadni & @tonault : After I decide whether or not to migrate the timelapse to HBR, I will give some attention to the GF2 as it should be quite simple. If you guys want to try it, the "meat" is here: http://www.personal-view.com/talks/discussion/comment/43237#Comment_43237 . If you want to do it on 720 just do it on the correspondent settings. It would also be of service if you guys could tell me what slow SS is available on the GF2 for video and which ones have a 360º shutter. You could share that info here: http://www.personal-view.com/talks/discussion/2396/gh2-motion-fluidity-360º-shutter-test-on-slow-shutter-speeds. And you don't need to compute the distances altogether as it's a lot of work, just check if the line on frame N+1 starts where the line in frame N ends and say Yes or No. You can see how I did it from the videos on the first post on that thread.

  • @duartix - "I believe 25p HBR still gets it's GOP from 24p."

    Yes, it does, but that's because both HBR 25p and 1080i50 get their GOP-length from the 24p setting. Likewise, HBR 30p gets its GOP-length from the 1080i60 setting.

  • Sorry for the lack of updates. The idea of using HBR for timelapses isn't dead, my time on the subject is being limited but I'm just doing a lot of related investigation in order to make it solid. This demands a lot more systematic testing, because HBR isn't straightforward like 1080p and this PsF thing is full of pitfalls, even accessing the IQ visually isn't simple any more.

  • @LPowell : I believe your " HBR is 1080i" sentence is starting to haunt me. :(

    My effort to port the timelapse settings to HBR 30p is facing what looks to be a dead end. Targeting 1/2.5s as Shutter Speed, I settled on GOP12 which should look like this:

    1 PPPPPPPPPPA 2 PPPPPPPPPPB ... (numbers stand for I-Frames, P,A,B,C for P -Frames)

    As soon as I trash P-Frame quality with a "1808i Scaling P" full of 0xFFFF, I'm getting something like this:

    1 pppppppppa 2 pppppppppb

    • On VirtualDub (through a DirectShow input plugin) the IQ is terrible with what looks like reduced resolution and persistent interlacing artifacts here and there. However it shows me updated content every other 12th frame as it should.

    • On @cbrandin 's StreamParser's viewer, the IQ looks good, but it's updating the video with changes every other 6th frame (as if I had wrongly doubled the GOP) but in fact I believe it's taking 2 PsF frames to build 1 Progressive (I say this because before the screen is refreshed I see a very blocky scene already with the motion updated just for an instant before it recovers some quality) but this is not reflected on the current frame cursor on the main window.

    • Playing on VLC I get a slight dose of shimmering interlacing artifacting (not so much on motion areas but mostly on high frequency static detail).

    This leads me to believe that apart from not having a tool that gives me consistent output, the content around point "2" is constructed from both P-frame "a" and I-Frame "2" which might be tragic since "a" is trash. This could justify the kind of IQ problems I'm facing and would totally defeat my strategy of trashing the size of everything other than I-Frames. I had a similar experience when I tried it on 1080i50/60 but I though since this HBR stuff was progressive capture from the sensor, it should also be progressive encoding.

    What do you think?

    P.S. Googling for "interlacing should die" shows: "About 2,330,000 results".

  • Have you (probably have) messed with the GOPx2time settings to halve the time, say 25000 for 50i?

  • No @driftwood, I haven't touched that stuff. :( I started with with @LPowell 's FlowMotion 1.11 and changed only "GOP Length for 1080i/p" and then I trashed the 1080i Scaler Tables for P. That's right after that that it happens. Should I start from scratch?

  • Well you should build up from your own philosphy first, perhaps its better than using anyone else settings. That way you methodically learn everything from Vanilla. There's plenty of people here who are right behind you and WANT you to get this working. Its a brilliant idea. Maybe share your latest settings as you go (here) so long time testers can provide help when coming to problems. It's your baby!

  • I'll start from scratch then, but I bet it's gonna happen again as soon as I trash the Scaler tables... I've got a feeling Panasonic was lazy and the encoder is interlaced (even if the capture is progressive)...

  • @duartix Trashing either I-frame or P-frame Scaling Tables in Flow Motion v1.11 will likely produce reliability problems in HBR mode. Each HBR keyframe is encoded as a pair of interlaced fields, using I-frame encoding on the top field and P-frame encoding on the bottom field.

  • Thanks @LPowell

    So basically you are confirming my projected worst case scenario. Do you think that should I enable B-Frames, that very same HBR Keyframe isn't built from a B-Frame also? I could do that and just trash quality on the B-Frames so I could recover bitrate/recording time?

    I've also tried twice playing with the GOP Table but I get absolutely no effect out of it... :( How's it supposed to work? I mean I tried 12,1,1,12,1,1, I've tried 1,1,1,1,1,1, even 1,0,0,1,0,0... NADA. Frame size doesn't change...

  • @duartix It's ok to trash the B-frame Scaling Tables, they aren't used to encode keyframes. If you do this, you will have only I and P-frames in your encoded file. I've experimented with GOP Table settings and concluded that it's not possible to patch the AVCHD encoder to produce anything other than IBBPBB... frame sequences when B-frames are enabled.

  • OK, thanks again, you are already answering questions before I ask them. :) I was wondering if I could enable even more B-Frames...

    Nick @driftwood 's take on the GOP Tables is that it is a sort of bitrate allocating normalizer, however I must be doing something wrong because I see no effect from changing it.

  • GOP tables are generally ok for 24p and 1080i if you use the stock GOPs, if not, you can change them but I think VKs code correctly does it automatically. However, if its 720 with B frames its all wrong. Let me give you an example; 12 GOP PAL 720 (changed from 24 default which is set for just p frames), fixed it should look like below (this is tested and Im using this in future Quantums);-

    12 GOP count should be 1 i frame, 3 P frames and 8 B frames;-

    72050opt1 = 1, 1, 4, 0, 2, 4

    72050opt2 = 1, 3, 8, 0, 0, 0

    Result. Clean b frame frames in proportion. elecard Stream analyser also now shows the B frame tables.

    Something can be said of the GOPxTime stuff too...

  • How does timebuster sit with quantum v9b or orion?

    What will it do? A mate said he had a patch that allowed for timelapse mode to only affect mpeg videos. I'm pretty happy with quantum v9b until further testing updates on orion occur, so i'm just after the timelapse mode, but not sure I understand how to get it, without affecting any 1080p24 modes.

  • I think this is the thread you are looking for: http://personal-view.com/talks/discussion/595/mjpeg-fps-setting-for-timelapse-on-gh2

    The disadvantage with MJPEG is that you can only get a max file size of 2GB before the camera will stop recording, thereby limiting your timelapse length. It might be long enough though depending on your needs.

  • @driftwood : Thanks Nick! When I find the time I'll try it on stock GOP. However, changing the stock GOP is 33% of my base settings :) and on the verge rising to 50% on HBR... :O

    @ricker : Timebuster Base will work fine if merged with quantum v9b or most of Nick's settings, however due to the nature of Nick's settings (which cap the Frame Limit to ~6000000) I recommend you choose the pack that has a higher "1080p24 Frame Limit" value such as FlowMotion or Cake and merge with one of them instead. You'll get less recording time (still over 24h) but you'll also get better IQ.

    Having said that, ATM Timebuster is all about getting very high IQ from I-Frames on 24p and forgetting about 24p for anything else. What you probably want is TimeHBusteR which will target the HBR modes instead, but progress on it has been painful and slow because of what's been discussed here, because I'm no master on AVCHD nor do I have aspiration to be, and because I have kept it to a very few settings that are mostly isolated from Bitrates, Buffers or most Limits, concepts within my reach but beyond my need to control/test.

  • Duartix, you might just forget about the idea of trashing the inter frames for HBR. The bit rate savings is small when you use a long shutter speed.

  • @balazer : I have to confess that between giving this baby for adoption, that idea has more than once crossed my mind, but I'm trying to explore all possibilities. The attached snapshot from StreamParser still shows that the bitrate is distributed as follows:

    I = 45%

    P = 30%

    B = 25%

    I'll be definitely trashing B (and even more eagerly P shouldn't there be all those shimmering issues), but there is another thing that is puzzling me:


    Frame--Type QP---Min---Max--Range---DC Skipped QST-High


    00013 ( I | P ) 20 18 21 4 6 434

    00014 ( b | b ) 22 255 0 -254 4064 0 *

    00015 ( b | b ) 21 255 0 -254 4064 0 *

    00016 ( P | P ) 20 20 20 1 6 2457

    00017 ( b | b ) 20 255 0 -254 4064 0 *

    00018 ( b | b ) 20 255 0 -254 4064 0 *

    00019 ( P | P ) 20 20 20 1 6 3262

    00020 ( b | b ) 20 255 0 -254 4064 0 *

    00021 ( b | b ) 20 255 0 -254 4064 0 *

    00022 ( P | P ) 20 20 20 1 6 301

    00023 ( b | b ) 20 255 0 -254 4064 0 *

    00024 ( b | b ) 20 255 0 -254 4064 0 *

    00025 ( I | P ) 20 18 21 4 6 435

    I-Frames are selling around the 130,000 mark at QP 20-24. How can I get them bigger? The bitrate use is miserable, shouldn't it all add up to just setting a big "1080p24 Frame Limit" and a low "Quantizer for 1080 modes" ? Or is anything else I should worry about?

  • fb2. have you 'worried' about that yet? fb2 should be an 8th of frame limit.

  • Thanks Nick, that's the kind of stuff that justifies a wiki.

  • @driftwood : Are you talking about 1080p24 FB2? Is it used for 1080i60? Any knowledge on why it is 1/8th? These settings of mine are rather unconventional and many times they defeat the logic of a normal AVCHD stream, that's why I'm asking.