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
  • Sorry, I meant the Quantizer for 1080 modes. I know about Initial Quantizer. But I've made a few preliminary retesting and it looks like it's coming from the GOP change. Stock GOPs on daylight footage honour the QP, the GOP changes I did for timelapsing are interfering with the QP, which seems odd since I'm using shutter speeds around 1/5s.

  • Hi folks! I've been out for a while and sees that there are new patches. Can someone make a summery of the benefits of using Cake rather than Orion?

    All I'm interested of is getting the highest quality in good light conditions.

  • David, the main difference between Driftwood's intra settings and mine are that intra is less efficient than GOP3 for the same quality, and intra has better frame-to-frame consistency. He also has some higher bit rate settings that are capable of higher quality.

    Intra-frame settings have an advantage over inter-frame settings for frame-to-frame consistency, because the GH2 encoder sets I-frame quantization parameters differently from how it sets P- and B-frame QPs. P- and B-frames will use the quantizer setting that you set, while I-frames will use that setting plus or minus two. The lower the quantizer setting, the less apparent the differences. At a quantizer settings of 22, the differences are small and not usually noticeable. These differences remain even if you change the scaling tables. (which I did, to make the P-frames more closely match the I-frames, but it was hardly necessary because the default scaling tables for P-frames and I-frames are nearly identical)

  • @davidhjlindberg, see the 1st post for Cake's mission statement.

    It all depends on your exact priorities - Cake and Flow Motion try to give good quality & consistency at reasonable bitrates & with reliability (across many recording modes) & spanning. They use low-GOP and are tuned for consistent frame-quality, but as a result they are also less efficient than long-GOP and you loose a little overall image quality. My (unfinished) Sweet16 attempts shifts the balance back to that quality, but probably sacrifices reliable spanning and low-GOP-style motion in the process. If you need top image quality above all else and have a fast/large enough card, you probably want Driftwood's high-bitrate intra patches (I've never tried them myself as I can't justify the file sizes & storage/backup reqs).

  • I am now touting the ability to choose your own quantizer setting as a feature in Cake. It was my priority from the beginning to have the rate control working properly and find the right limits for reliable operation, so that you could use any ISO and quality level that you want. I chose a quantizer setting of 22 simply because it offered a balance of file sizes and quality that made sense to me. Someone else might want higher quality and larger files, and with Cake, that's easy to do: just change the quantizer setting to 20 or 18. Below 18 you're going to be hitting the limits too frequently, and so you should look at settings with higher limits, such as bkmcwd's Gop3zilla.

  • @_gl Good summary of the trade-offs of different patch optimization strategies. I'd also add that my research on Flow Motion confirmed one of the claims made by Intra-frame advocates:

    The image quality of P and B-frame macroblocks produced by standard AVCHD encoders is inferior to the image quality of I-frames.

    That's because the built-in Scaling Tables in standard AVCHD encoders use coarser quantizers in P and B-frames than in I-frames. To fix that problem, I customized the Scaling Tables in Flow Motion v1.11 to produce P and B-frame image quality comparable to I-frame macroblock quality. This allows Flow Motion to maintain consistent I-frame quality levels while reaping the efficiency benefits of short-GOP encoding.

  • Thank you all, really helpful information!

  • I've been using Cake for quite a while now, and had a dabble with the Sedna settings but to be honest I prefer cake and have gone back to it - for the same reason that @_gl sites, that the new settings are too hungry on card space for the cards / use I need, which is shooting 4-5 minute sequences of music. Cake seems to do what I want particularly with QP set to 20. Also it's nice to stick with a setting and get used to what it does rather than switch around constantly trying new settings (I think).

  • This was going on outside my window one morning - the steaming trees of death!

    For ipod users it's at http://vimeo.com /38840506 (remove the gap)

    Cake 1.2 QP20 plus a bit of curves in post. Last shot ex tele mode. All 14-140 lens. I lost a bit of my tripod so couldn't use it - then found it just after all the mist went away.

  • @Mark_the_Harp, v. cool. What caused that?

  • @_gl It was after a cold night's frost, and the sun basically hitting all these surfaces and the frost evaporating. It lasted about 10 minutes, and I was lucky to catch it just as the sun was getting onto the garden. Plus it was a very still day and the light was shining in the right direction to capture all this. The foreground steam is rising up from the roof just beneath my window.

    I thought it was quite fun to be able to shoot a whole sequence of stuff through just one window!

  • New: Cake v2.0

    • Designed for reliability

    • All modes

    • 2 - 2.5 times stock bit rates

    • Rate control by adaptive quantization

    • Spanning in HBR (NTSC and PAL), 24p, and Variable Movie Mode

    • SanDisk Extreme cards recommended for spanning

    The settings are attached to the second post of this topic.

    Cake v2.0 has reliability and spanning as its highest priorities. It's intended for event coverage, where you need long recording times and reliability. My testing is based on the most extreme shooting conditions, and I've set the bit rates very conservatively to ensure consistent spanning with SanDisk Extreme cards.

    Special attention was given to HBR mode, to keep its reliability and quality high. Last I checked, I have the only stable high bit rate inter-frame settings for HBR.

    Of course quality is still a priority, and these are my best looking settings yet. Rate control is by adaptive quantization, very similar to how the stock encoder settings work. When operating at the bit rate limit, adaptive quantization has better image quality than do settings using fallback mode or frame limits.

    Cake v2.0 is my attempt to be done with the testing and the pixel peeping, so I can get on with the shooting and never have to worry about my encoder settings not being up to the task.

  • Holy shit!

    How did you manage to work it that stable at QP=18? It's the GOP table trick? IIUC you are giving equal bitrate to I/P right? How did you enable adaptive quantization? Which matrices are you using? Sorry for all these questions but I'm still longing for the wiki...

    Thanks! Good work!

  • How did I make it stable with a quantizer setting of 18? Basically, I didn't change any settings except for the ones that I understand and needed. ;) Changing GOP tables is how I got adaptive quantization rate control to work at the desired bit rates of each of the different modes. That was independent of the quantizer setting. I-frame quality is matched to P-frame quality as best I could make it, same as with previous versions. The I-frame and P-frame bit rates are both adaptive. Under high noise or high motion, the P-frames will be nearly as large as the I-frames, to achieve the same quality. Under low motion, the P-frames can be much smaller, still with equal quality.

    P-frame scaling matrices are copied from the stock I-frame scaling matrices. This wasn't strictly necessary, as the stock P-frame scaling matrices are already very close to the I-frame scaling matrices, but I did it anyway just so they'd match.

  • Cool - looking forward to trying this once my camera is charged up. I'm a big cake fan!

  • OK, I'm flashing now with Cake 2.0, exciting ...

  • Just checked, I and P frames are there , @balazer is there a reason no B frames ? It seams to work ok with HBR 25 PAL ... I think this will stay on my GH2 for the while :-) Thank you

  • I am liking this cake more each time I try it..trying Cake 2.0 now...Video to follow! Thanks Blazer!

  • Thanks for the feedback, guys. Crossing my fingers that no one will be able to make these settings crash. Do your worst. ;)

    Feha, I just see no advantage to using B-frames. The GH2's B-frames, from all my testing, are not much more efficient than its P-frames, and the P-frame quality matches the I-frame quality very closely even without changing the scaling tables. P-frames always worked well for me through the iterations of Cake, so I just kept them. P-frames are simpler, so they might be helping the stability.

  • @balazer, ok will do some testing tomorrow and see how it holds, the bit rate is great ... I'm using Extreme HD video 32GB / 30mb/s hope it will span ...

  • Got to the same P/B conclusions in my Timebuster settings. P were more efficient, even though my settings were rather unconventional...

  • A quick test of cake 2.0...Not too scientific ...I am new to this camera and the hacks. Windy overcast day here,so light was not the best. Some shot in Smooth -2-2-2-2 and the rest vibrant. I did try the 720 60FPS and looked fine to . Cant wait to get it out with some good light for contrast comparisons. I did try the Sedna before this but had it stop with the 720 60fps and the files were huge !..Nice for 1080 24p though. Using a Sandisk extreme 64g 95mbs .

  • Looks nice, thanks for testing. I see some blended frames. How are you editing/converting?

  • I use Sony Vegas...Could you explain the blended frames you see?...I am pretty new to this.

    Thanks!