Personal View site logo
Make sure to join PV on Telegram or Facebook! Perfect to keep up with community on your smartphone.
Please, support PV!
It allows to keep PV going, with more focus towards AI, but keeping be one of the few truly independent places.
GH2 - 2.5fps AVCHD 24h Timebuster/TimeHBusteR 2.0 settings - The day is not over!
  • 323 Replies sorted by
  • @duartix it is strange that one of the first frames has to be deleted to get the right distance of 10 frames at 24p 2.5fps. Do you have any idea if spanning takes place and subsequently files were put together that I-frames are always resulting in the right distance of 10...

    How could this be checked...

  • Use after effect to speed it up work great for me with Gh2. Do you think timebuster work with Gh1 also?

  • @tinbeo : For it to work the following settings have to be supported on GH1:

    • 1080i50 and 1080p24 GOP Size or 1080i60 GOP Size
    • Encoder setting 1 1080i/p
    • Quantizer for 1080 modes
    • 1080p24 Scaling P/B or 1080i Scaling B

    Load up a GH1 fw in PTTools and see if those are available.

  • @tida I don't know why that is, but it's probably because of that rogue P-Frame that appears in the first GOP. In a lot of merges I've observed the following I-Frame cadence:

    • 0, 11, 21, 31, 41, 51, ...

    so the GOPs are:

    • 11, 10, 10, 10, 10, 10, ...

    and that is the reason.

    From what I've observed this ONLY happens in the first GOP. To test it I guess that opening the second file of a spanned set in StreamParser and looking at the first GOP structure is enough.

  • @duartix I thought that might be cutting first file at the end and subsequent at the beginning by smartcutter or other non re-encoding cutter programs will give a hint. In addition merged file cut within the target area. Hope streamparser will read it correctly.

    I'll give it a try when I have time for it...

  • Just analyzed one video which spanned twice and looked at all three files if they would be put together correctly. I took the video at 24p and 2.5fps and hence @duartix timebuster GOP 10 settings.

    I did not load resulting long files into streamparser directly as it might take hours or weeks of reading time. Hence I did cut them by frame accurate smartcutter program.

    I saw that first and last few frames of files are modified by smartcutter but numer of frames from the start are kept correctly in place when I analyzed them by means of streamparser.

    I did cut all files few seconds after the start as well as few second before the far end.

    • Result of 1st file at start: I-frames cadence at 21,31,41 etc..

    • Result of 1st file at the end: after last I-frame 9 more small frames

    • Result of 2nd file at start: I-frames cadence at 20,30,40 etc..

    • Result of 2nd file at the end: after last I-frame 9 more small frames

    • Result of 3rd file at start: I-frames cadence at 20,30,40 etc..


    Good news for 24p at 2.5fps !!! There seems to be no hidden issue. Extraction of I-frames can be done correctly. Just put spanned files together, skip 1st frame and choose every 10th frame subsequently.

  • I have never done timelaps before but this looks easy enough that I will have to try it out. @duartix, yes I say include the AE workflow in the Timebuster documentation.

  • @tida: Good to know! I haven't yet tried importing spanned files on AE but your test results are painting the best picture for simplifying the workflow, be it concatenating first and importing later or importing bit after bit.

    @HillTop1 Will do!

  • Please find attached Avisynth script which would work with 24p/2.5fps spanned files correctly. Just put into the field of entries your target frame rate and timelapse rate.

    Timelapse 24p 2.5fps.txt
  • @duartix, do you think it is OK to use 24h Timebuster Base merged with new Flow Motion v2.01? I don't really know how combining settings affect each other..

  • @vstardust

    I do. If you peep the file, you'll see that 24h Timebuster Base is dramatically simple and was designed to merge without adverse effects.

    I was planning on releasing an updated pre-merged version since Flow Motion 2.0 was out, but there was quite a share of error reporting on it and I was waiting for the dust to settle, be it in current Flow Motion 2.01 or 2.0x... If things stay quiet, I'll do it in a few days.

    In your case it's just a question of downloading FM 2.01 and 24h Timebuster Base. First you load 24h Timebuster Base on PTTools, then you “Alt + Click” on FM 2.01. It's ready to save & flash.

  • @duartix Yeah, I tried it with Cake 2.3 yesterday. Merging worked with no problems. I guess the 24p timelapse footage is supposed to look f*cked up before time remapping because P-frames and B-frames are trashed.. but the first time I saw it with VLC player scared me a bit :) One more thing - are there any good alternatives to Virtualdub for Mac? I don't have Adobe AE either right now.

  • @vstardust

    Yes, fucked up frames are a normal consequence of these settings, but you shouldn't worry because there should one pristine I-Frame in each GOP / unique content capture. AE workflow is really the simplest I could access. I'm on Windows 7 so I can't tell you much about Mac workflow except that VirtualDub could possibly work under some kind of emulation. @tida has worked on some great scripting for frame weeding but I don't know how AviSynth works under emulation either.

    I guess your quickest assessment is AviDemux. It looks like it has a Resample Fps filter (you should uncheck the Blend box because you don't want bad frames to contribute to the output) but I don't know how easy it is to control which frames are chosen (possibly it's done with the markers).

  • Since FlowMotion 2.02 looks very stable now, I'm releasing 24h Timebuster/timeHBusteR pre-merged versions making it easier to those who are not comfortable merging settings. There is nothing new regarding timelapsing except for FlowMotion's 2.02 changes.

    • 24h Timebuster 2.02 FlowMotion which is 24h Timebuster 2.0 Base merged with FlowMotion 2.02

    • 24h timeHBusteR 2.02 FlowMotion which is 24h timeHBusteR 2.0 Base merged with FlowMotion 2.02

    (Files are at the OP: )

    The day is not over. :)

  • @duartix thanks, that's great. What is the difference between Sanity 5 and FlowMotion at 24p. Would it be possible to get the maximum I-frame quality if shutter speed is set to 2.5fps (GOP10) for timelapse as well as fairly good video results if it comes to >25fps.

  • @tida : @Ralph_B and @LPowell could tell you better, but from a what I've gathered, Sanity is all about efficiency, which I believe it to be an undisputed champion. For that purpose it uses long GOPs. FlowMotion OTOH is a very high quality/high bitrate setting. IIRC in it's latest incarnation, Lee has broken new ground with the tuning of Deblocking Filter parameters. My excuses to other developers if this isn't true. The maximum I-Frame quality usually depends on a combination of the 24p Frame Limit (Higher is better, look around in the thread because I believe I've talked about it in here before) and the Quantizer parameter which is also a tell sign of how much the I-Frames can be degraded (in this case lower is better). You'll have to look at both settings and see what their values are. I can't look at those right now, I should have been to sleep one hour ago, my head is about to explode...sorry!

  • Has anyone else had problems working with files from 2.02 in AE? At HBR settings, the frame rate (as read by AE) is 29.985, totally freaking AE out giving an error message "overflow converting ratio denominators" so the files are unusable in AE. If i interpret the footage as 29.97, the clip shrinks from over an hour to around 20mins.

  • This could be due to the timings in time_scale (set in GOP Related x2 time) or x2 GOP. I believe these dont really need to be changed as VK's ptools correctly works this out (or isnt really useful as of yet). Correct me if I'm wrong @cbrandin @Vitaliy_Kiselev.

  • I'm sure for a lot of people all this tech talk is way over (my) head. Is the main advantage of all this work to extend recording times for super long timelapses? For an hour or two of timelapse is it possible to use the 300% vmm with a shutter speed of 2 with any good patch?

  • @mrttt: Yes! and It depends!

    GOP = a Group Of Pictures, is like this: I P BB P BB P BB P BB

    • I = Independent (like a photo of the footage)
    • P = Predicted (what changed from the previous frame)
    • B = Back predicted (same as P but can describe changes from the next frame too)

    Let me start by answering your second question first.

    I'd say that 3/4 of the performance gains of Timebuster rely on marrying the GOP with shutter speed. When you are using a 1/2s shutter speed, you would only need to record 2 different frames per second because all the other frames are just duplicates of the same content. Using a GOP1 Full Intra setting (one that records every frame independently - exclusively I frames) is massively inefficient when compared to using a GOP13 setting (one that records one frame (I)ndependently and the next 12 frames as the changes from that first one). Since those next 12 frames have the same content, they can be easily encoded with a very small difference frame (P)redicted or (B)ack predicted that just says "Nothing in the frame is new and nothing has moved" instead of enconding the same whole picture again.

    What does this mean? It means that a good patch with long GOPs will be efficient in timelapsing because P/B frames describe the content repetition with a whole lot less space. It will be optimal if the GOP is aligned with the shutter speed resulting in one I frame for new content and all P/B frames for repeated content. This kind of approach should give you ~8-10x less bitrate than using a Full Intra.

    Now, the other 1/4 of performance gain results from further squeezing B/P frames to an absolute minimum without caring for IQ because since these frames will not be used for the final output, they can be trashed to oblivion.

    Finally the first question. The consequence of these settings are ultra low bitrates so: Yes, the recording times are massively extended BUT, they were designed to be flexible in the way that they allow you to opt for IQ instead of recording length: all you have to do is lower the Quantizer Parameter and you could have the same recording length but at much higher image quality levels. You've got that choice. :)

  • I used mainly Timebuster and Sanity settings for my hacked GH2.

    My DIY motorized slider gives me actually 8minutes for a shot in slowest mode. You will see shots with lenses like Lumix 14-140mm, Olympus 45mm/1.8, Lumix 20mm/1.7, Canon FD 35mm/2.0 FD, Canon FD 135mm/2.8.

    Workflow by AviSynth/AvsPMod/MEGUI. Takes no longer than 10 minutes to prepare a file even from a 2 hour shot - including rendering with i7/2600K.

    Special thanks to Vitaly, Duartix and Ralph_B

  • @duartix

    Can you summarize the workflow now with the new information on SS and the latest version of your patch, or is your April 18 post above the latest and most accurate instructions?

  • I think they are still current, works well for me.

    The only point still a bit unclear is the begin of sequence (first I-frame) when using GOP 5.

  • @ethanpil go to the first page which was always updated by @duartix.

  • @tida I guess its still confusing for me.

    If I simply upload the 24h timeHBusteR 2.02, what do I need to do to my camera? Manual mode... Do I use 24H recording mode?

    I read about ISO 160, do I need to set that before recording? What about the VirtualDub steps mentioned on page 1?