Personal View site logo
GH2 Cake v2.3: reliability and spanning in 720p, HBR, 24p, and VMM at 2-2.5x stock bit rates
  • 609 Replies sorted by
  • @duartix. Ok, I will try. As for the difference between hbr and 1080i25p, I was able to check on my tv. HBR appears to be flashier, more video. Am I right or is it the vision of a novice.

  • Thanks for the report, oronce. You seem to be confusing the mode names. Tell us exactly what you set in the menus, e.g. HBR PAL, or 1080i PAL FSH.

    Duartix, it depends what you mean my stress. High detail that continually changes will induce high bit rates or QPs. Modest motion can cause unpredictable frames sizes.

  • @balazer. I tried two settings. 1st Creative mode on Manual movie mode, then on Motion picture: Rec Mode AVCHD (1080i) and Rec Quality: FSH. 2nd mode is HBR (don't have to select anything on "Motion Picture) Both are seen by StreamParser as 1080/50i. My GH2 is a pal one. Hope it's clear. About stress it appears on my last test that the bitrate was the highest as i was shooting in an almost completely dark room with almost no motion. Is bitrate meaningful in term of stress? Sorry for the candid question.

  • Don't worry about stress. The encoder with these settings should handle whatever you throw at it. 1080i (FSH) will look like video due to the field rate of 50 fields per second. HBR mode will look more like film, because of the frame rate of 25 frames per second. Also, HBR is a lot sharper on the GH2 than 1080i is. 1080i needs to be displayed as interlaced, or de-interlaced for display progressively. If you just want to play files on your PC or upload to YouTube, HBR is the better choice (and switch from PAL to NTSC in the setup menu so you get 30 fps instead of 25).

  • @balazer. Thanks a million for the info and for the work. 12 years ago I stopped working as assistant director on feature films and my last short movie is 20 years old and shot in super 16, so video is very new to me. I am planning to shoot a documentary in pal, so I will set it HBR pal. Thanks again.

  • For a documentary, I'd first think about intended screening. If aimed at (HD)TV, I'd rather go 1080i50 or 720p50. For the internet or the big screen it's 30, 25 or 24 (all progressive).

  • @balazer. Haven't tried Cake 2.0 because I use 1.2 and love it. Is 2.0 better for 24p (I'm talking image / motion quality here)? And if so, how exactly? Looks like the bitrate is lower in 2.0 so that's why I ask.

    I would like to settle on a good setting, because at some point you have to stop trying these different settings and actually use them for real. For me, Cake 1.2 is the one I've been really happy with and the one I always return to.

  • Thanks. I think I got it.

  • Thank balazer for creating the most ideal patch so far!

    Inspired by your dynamic quantizer, I've created such a patch for my GF2 (I don't have a GH1/2). I don't like GOP 3, so I use GOP 6 and make the B frames as good as I and P like LPowell's Flow Motion (another most inspiring patch so far). To make 1080i60 absolutely stable (panning quickly from a very dark area to a very bright and detailed area such as pavement), I have to change the GOP table from:

    10 20 20 10 0 0


    10 20 20 10 10 10


    8 16 16 8 8 8

    What's interesting is: not just no more crash, the video look better with more contrasty (or darker) and sharper. I like this result, just wonder why. Balazer and others, could you please explain?

    My other key changes:

    FSH/SH Bitrate: 99000000 All Top Setting: 90000 All Bottom Setting: 80000 All Scaling Fallback: same as I/P/B

    The idea is to make no any other bitrate limiting other than GOP table. The average peak data rate is about 48Mbps, absolutely stable so far, no failure yet even at the most difficult conditions where failed before, 4GB spanning fine with a 32GB Samsung microSD class-10 card. Even class-4 cards work fine, just no spanning.

    I've tried 44000000 AQ4, 66000000 QP 30 worse B-frame and QP 35 same I/P/B and played with many other settings, nothing compares to this GOP table method.

    My GF2 is now like a new born. Clicking on the donate button... few dollars make my $200 camera a $$$$ video gear, the freedom of configuring my own device and so much fun!

  • @pvdog care to attach settings? The GOP table changes make sense to me, but increasing the FSH/SH bitrate from 55000000 to 99000000 to gain stability on the GF2? (almost a x2 increase)

  • Mark, yes, Cake v2 really is better for 24p than previous versions are. I've lowered the quantizer setting from 22 to 18, which improves image quality when the video can be encoded at the lower quantization level within the bit rate limit. When the encoder operates at the bit rate limit, adaptive quantization looks much better than fallback mode. I-frames and P-frames will still be encoded with equal quality and maintain the ratio of frame sizes (I-frames larger than P-frames by some factor). Fallback mode doesn't do that. Using fallback mode for rate control was a nice trick, but only since I didn't know how to make the rate control work properly at the time. Even with v2 operating at slightly lower max bit rates, the picture looks better with the different rate control scheme. The difference is not huge in 24p, but it is huge in HBR. All of this work was motivated by HBR, because it simply did not look good before, and now it looks very good. I know people have settings fatigue, with so many different choices and frequent changes. I went out of my way to make sure that v2 would be rock solid, and so far at least, I can say I have not had a single write error, lockup, or spanning failure, and no one has reported any. I hadn't with v1.2 in 24p either, but I think spanning failures were a possibility with the way I had set the bit rate limits. Fallback mode is slow to kick in. Several GOPs will pass before it starts to limit the bit rate, and before that happens, with the GOP length and quantizer setting I'd used the bit rate could go as high as 100 or 110 Mbps, which would be enough to cause a spanning failure at GOP3 using an Extreme card if that were to happen at the moment of spanning. I did experience spanning failures in HBR mode in Cake 95. Adaptive quantization reacts very quickly to bit rate changes, setting the quantization level for the next frame after a change in the bit rate of the last few frames. So I can get closer to the spanning limit using adaptive quantization than with fallback mode. But even with adaptive quantization's fast response, the camera can't predict how large the next frame will be when coded with a certain quantization level, so you can still encounter bit rate peaks above the max average bit rate. But those peaks will be very short - generally limited to one GOP, and I've set the bit rate such that the highest peak I'd ever seen in all of my testing is still low enough to span.

    I don't anticipate making any more changes to 24p, VMM, 1080i, and HBR. I do plan to go back and make 720p span.

    PVdog, I don't have a GF2, so I can't say exactly what is happening there. Increasing the GOP length will decrease the maximum average bit rate. Increasing the values in the GOP table will decrease the maximum average bit rate. The values in the GOP table are counts per half second of the following:

    1. I-frame (progressive) or I-frame first field (slice 1, intra coded)
    2. P-frame (progressive) or P-frame first field
    3. P-frame second field
    4. I-frame second field (slice 2, p-coded)
    5. B-frame (progressive) or B-frame first field (slice 1)
    6. B-frame second field (slice 2)

    If you use a GOP length different from stock or a mix of P- and B-frames that is different from stock, the encoder will not use the stock bit rate. Changing the values in the GOP tables is a way to compensate for that and get whatever max average bit rate you want at any particular GOP length. I try to keep the GOP table values in the ratio of actual frame/field counts per second of video. The Top bit rate setting should then be set to the max average bit rate you've achieved, or slightly below it, and the bit rate setting and Bottom bit rate should be set relatively.

  • Excellent, Balazer, thank you so much. I'll definitely give it a go - my battery is loaded and the ini is ready to put into ptools. Thanks! There is a sort of "settings fatigue" because often the pictures don't look that different between settings (they're all very good). So if I'm drawn to return to a particular one (in my case, Cake) then it's probably because it really does suit the stuff I shoot and the settings / post-pro I do. It just "sits" very well and I find it easier to know that it will look good (if that makes sense). So it's useful to know what went into Cake 2.0. Thank you - I'll have a go with it!

    Cheers...and thank you also for all your work.

  • @balazer : Thanks for the explanation. Very clear and useful in wiki challenged times.

  • Thanks balazer, now it makes sense. Somehow I read in this forum that the order of GOP table was like this: I-eve, P-even, B-even, I-odd, P-odd, B-odd; that was why I had no idea about what was going on.

    Based on your description, there seems an error in your ini file:

    720p60 Opt2 GOP Table=12, 60, 0, 0, 0, 0 720p50 Opt2 GOP Table=12, 48, 0, 0, 0, 0

    60 and 48 should have been swapped.

    I'll start a new thread to make it bold for GF2 owners' attention.

    Attached setc.ini is the stable one fully tested; sete.ini has slight changes (from 10 20 20 10 10 10 to 10 20 20 10 15 15) based on above balazer's explanation with 90/72 GOP length. The latter should be even more stable as bitrate is lower than the former (still much higher than balazer's), but I have not tested it much yet.

    The following clips were shot with the 10 20 20 10 0 0 GOP table, it was not fully stable and I lost some very cool shots. I tried all methods such as narrower top and bottom and lower total bitrate as balazer suggested above, but could not prevent occasional crash until I changed 0 0 to 10 10.

    I had to re-compress the video to reduce the size for easier uploading; so what you see can be only even better. These are simply the very best video from GF2 I've seen by far. I believe the new 10 20 20 10 15 15 should be very close in terms of image quality and motion smoothness.

    Thank VK, I've donated, will donate more if I'll get a GX1 or GF3.

    I started a new thread to get GF2 owners' attention:

  • pvdog, it wasn't a mistake. My settings are GOP6 for 720p/60, which makes the ratio of I-frames to P-frames is 1:5. It's GOP5 for 720p/50, so the ratio of I-frames to P-frames is 1:4. I'm not certain if the ratio really matters, or just the totals or just the P-frame values. I just like to keep the ratios correct, and it did what I wanted. I believe the Opt2 GOP tables are for 720p using P-frames, and the Opt1 tables are for 720p using B-frames.

  • I see. If you look at the stock values, Panasonic is apparently always trying to make them 30 or 24 total: 1 9 0 0 20 0; 1 7 0 0 16; 1 28 0 1 0 0; 1 22 0 1 0 0.

  • @pvdog Amazing, i must get an GF2 as B camera, great video, high quality ...

  • I am looking to do long records with 720p on the GH2. I have no real clue about spanning and whether it is an issue at this resolution. I will be shooting 15-20 minute takes. Will Cake do the business for me here? From what I have read 1280x720 50p doesn't get a lot of attention when hacks are formulated. And it seems Cake is more focussed on HBR and 24p

  • @jam: "Be careful that you reload the original firmware or reset all options before you run PTools"

    How can options be reset before running PTool? I have to reset options >within< PTool and therefor it has to run already. Or do I get something wrong here?

  • I have some SanDisk Extreme cards (20 or 30 MB/s) but also a couple Class 10 Transcend cards. Is Cake going to be okay on these Class 10 cards?

  • i have trasender class 10 and dont have a problem with this

  • @rsquires, I just shot an long interview (2 GH2s and AF100) using the Cake 2.0 patch and I had no problems with spanning using 1080 24p. From what I have read, 720p can be tricky and haven't tried using this patch in that mode.

    I've also used Lee Powell's Flowmotion v1.11 patch and it spans fine, again in 1080 24p as well as Driftwood's Dark Matter V3 patch.

    I'm using the 95mbps 64Mb SanDisk Extreme cards, though, as the slower cards were hit or miss.

    To be safe in all modes, I understand that @Mpgxsvcd's No Adverse Affects patch (42mbps) is probably the best route to go if you are doing long takes.

  • Other Class 10 cards should work, but I only guarantee spanning with SanDisk Extreme cards.

    rsquires, 720p H mode probably will span, and SH mode probably won't.

    sir_danish, to reset settings in PTool you uncheck all of the boxes. It's easier to just reload PTool.

  • @sir_danish Sorry to be unclear. Reload original Firmware 1.1 or uncheck options. What I believe I did was load one ini file in PTools and then another resulting in mixed settings, yielding some astonishingly bright strobing at highlighted edges.

  • Cake v2.0 1080 24p using Nostalgic -2 X 4.

    AF100 was main camera (working on a complimentary profile for the GH2) using an Olympus 14-35mm f/2.0.

    GH2 lenses were 25mm Voigtlander f/.095 (filming from the left) and LOMO OKC1-50-1 f/2.0 (filming from the right.)

    3200K white balance, 200 ISO.

    Slight color adjustment, Spanned fine using SanDisk 95mbps 64Gb cards.