@balazer - Are you familiar with the "b-frames" work of 'pvdog'? Pvdog claims that if he combines your Cake 2.3 with better b-frames (taken from FlowMotion) he gets the best of all worlds. His "pvdog" patch has been tested a lot on GF2 and also GF3. I have used it on GF2 - very nice. Does Cake 2.3 now take all this into account about "FlowMotion-type" B-frames? Would love to use it on GH2 if it does.
I look forward to your reply.
I made the decision to use P-frames and not B-frames because it gave me high reliability in all modes and the behavior of the rate controller that I wanted, with only a slight reduction in efficiency. I wasn't getting the same behavior from the rate controller with B-frames.
@Balazer - can you please comment on what PVDOG says about this? Here it is-
"This patch is inspired by two GH2 patches: LPowell's Flow Motion and Balazer's Cake.
The key features are:
1) All I/P/B frames are the same quality. You know, Panasonic uses worse quality for P and B frames.
2) Use GOP tables to control max average bitrate, much better than frame limit and fallback etc., for optimal quality and stability.
3) GOP 6, I-B-B-P-B-B-I. I always feel 15/12 is way too long but 3 is too short and cannot take the advantage of B frames.
4) Reliable 4GB file spanning.
5) Video plays fine in camera.
6) No compatibility issues with video editing software.
Fully tested, no crash yet even under the extreme conditions such as fast panning from very dark to very detailed, very bright areas. Average bitrate is about 45Mbps, Class-4 works fine, but class-10 is recommended for reliable 4GB spanning.
Attached setc.ini has been fully tested, not a single failure; sete.ini has some small changes for more standard 90/72 GOP length, should be even more stable.
Image quality and motion smoothness are simply superb, incredible for a $200 camera, I've never seen such sharp video..."
What are your thoughts on this, Balazar?
It's not the approach that I would take, but that's not to say it's bad. You can try it and compare.
@Balazer - The reason I am asking is because I am about to start shooting a full feature film. I may use Cake 2.3 for it. But I would love to learn about this aspect first. Any comment on PVDOG's reasoning (below) for the way he did things?-
"There are two facts about my comment on "same quality for I/P/B..."
1) If you look at the ini file or PTool, you can see: Scaling for I/P/B and fallback all have the same value. Stock values are worse for P, B and fallback. So my patch ensures constant good frame quality for all GOP frames and always - no fallback. I learned this trick from LPowell's very new Flow Motion patch.
This is also why I'm not interested in the AQ4 patch (All to detail), because it is long GOP with lots of P and B frames and I don''t know if it still uses the stock coarser scaling for P/B frames and fallback.
2) If you look at the GOP tables (under GOP Related), you can see: the stock values for I is much lower than P and B, putting a lot more weight on the coarser P and B frames. For example, 1 for I, 4 for P and 10 for B, i.e., P gets 4 times of weight over I and B 10 times. My patch lowers those ratios to 2 and 1.5. This trick is recently discovered by Balazer in his Cake patch.
We all know, I frames are the most important and they should take up more part into a GOP. More I frames mean larger file size and faster cards, but 32GB class-10 cards are extremely cheap now for only $17, can still give you 2 hours non-interrupted recording even after this patch!
Basically my patch combines all the latest great discoveries by LPowell and Balazer together with my own tweaks, it should be the very best by far, for both GF2 and GH2."
Please comment on this, Belazer?
The reasoning is all fine. There's just a huge amount of subtlety in designing settings that goes way beyond any of this reasoning. I chose P-frames because the GH2 rate controller sets the P-frame and I-frame QPs in the same way. I chose a GOP length of 3 because I could get the bit rate higher that way and still ensure spanning. The increase in quality from using a higher bit rate more than made up for the loss in efficiency in using a shorter GOP, especially at higher ISO settings. At lower ISO settings, there might be no difference, or there might actually be an advantage to using a longer GOP. But the quality is already more than good enough at lower ISO settings.
It's not to say that my approach is any better or worse than anyone else's. I had a set of priorities, and I made certain decisions that led me down a certain development path. What I have now works very well, and I'm not about to revisit those decisions given the huge testing effort that any change requires. There are better ways of improving my video than to endlessly tweak and test the encoder settings.
It's kind of silly to talk about the merits of different technical approaches if you haven't actually tried the settings and compared the results. I think you'll find that my approach works very well for the vast majority of shooting situations.
Thankyou, Balazer. I truly appreciate it. Tell me - Is the "2.3" download in the second post of this thread - is that the newest one which incorporates the "FlowMotion" influence? ie. The latest of the latest?
And lastly, when you use it with 24p, what is the "average" bitrate usually? 45 Mbits or so?
"For example, 1 for I, 4 for P and 10 for B, i.e., P gets 4 times of weight over I and B 10 times."
There might be another reasoning behind this. Don't forget that the stock GOP consisted of 8 B-frames, 3 P-frames and 1 I. I was under the impression that these GOP tables were simply used in the bitrate allocation mechanism and had nothing to do with the quantization, which is the parameter that really defines IQ for each type of frame.
@balazer - I truly appreciate your comments. Tell me - Is the "2.3" download in the second post of this thread - is that the newest one which incorporates the "FlowMotion" influence? ie. The latest of the latest?
And lastly, when you use it with 24p, what is the "average" bitrate usually? 45 Mbits or so?
@AndyS Dude, just try it and check how it works for you. Hope you didn't want to start shooting your full feature film before trying the patch in detail? I like this patch a lot, but it is no guarantee it will serve you as good.
@tetakpatak - Thanks Tetak. Yes - I will. It seems to me the best option as far as I can tell. Up until now I have just used Hacked GF2 - but just got the GH2 for this movie project. Does anyone know if the "2.3" patch in the 2nd post on this thread is the absolute "latest" one?
Thanks! - Andy
I already answered your question via private message. Check your inbox.
Hi, @balazer!
A question regarding spanning, this is what still says in your first post:
I recommend the SanDisk Extreme Pro 32-GB 95 MB/s card. (not 64 GB)
For some reason I haven't noticed this remark, so I've used the Cake 2.3 for months with the Sandisk 64GB card - but it just always worked perfectly well. I did many very long takes and the spanning problem didn't occur at all- not once.
Please explain that remark why not 64GB card, or is it outdated? If so, maybe update your first post and I will delete this comment afterwards also - to avoid any (imaginable) confusion.
which film mode do you recommend to use with cake v2.3?
I don't recommend 64-GB SDXC cards simply because of the "file number limit exceeded" problem that can happen probably with all SDXC cards in the GH2, hacked or not. (see http://www.personal-view.com/faqs/gh2-hack/gh2-hack-faq ) Aside from that, I expect Cake to work fine with the SanDisk Extreme Pro 64-GB card.
I have no film mode recommendation for Cake. The encoder can handle whatever you choose.
Done with Cake 2.3 24H setting
Cake HBR Test 1:
HBR 30p mode GOP5 at 65 Mbps; not spanning
HBR 25p mode GOP3 at 61 Mbps; not spanning
SanDisk Extreme Pro cards recommended
24p and Variable Movie Mode same as Cake 2.3 (high reliability and spanning)
720p H mode same as Cake 2.3 SH mode (high reliability and spanning). Don't use SH mode.
This was a small experiment to maximize the quality of HBR 30p mode, especially for dealing with the noise of ETC mode. I gave HBR 25p mode a boost also.
HBR 30p spanning is not reliable, at least not with my cards. It might span with an Extreme Pro 64-GB card.
These HBR 30p encoder settings are more aggressive than in regular Cake. Write errors and lock-ups are possible, though in two months of use I had just two lock-ups. Using a Lumix lens slightly raises the chances of a lock-up. I did the bulk of my testing with manual lenses.
I've tested ETC mode with these HBR 30p settings against top settings from other developers, and mine compare favorably, with higher bit rates, higher quality, and more consistent frame-to-frame quality.
Thanks for the patch. Shot a bunch of interviews this week with Cake 2.3 at 24p. It all looked great and no problems at all with Extreme Pro 32GB cards.
@balazer
Although you don't recommend those cards, I find it important to mention here that using your Cake v2.3 with Sandisk SDXC 95MB/s 64GB card gives as result the most reliable spanning imaginable.
From my own experience I can say this patch is fantastic and it is pity it remains somehow underrated among many good patches for the GH2. But there are no so many patches with such high IQ in combination with so reliable spanning in 1080p24.
The bitrates (many people seem to find it so important) with Cake 2.3 are also very high, in high detailed scenes are around 80-90Mbps average, so that my old PC system can't handle it. It is far higher than 2-2,5x stock bitrates.
Thanks for the feedback and praise.
The encoder settings span equally well with any type of SanDisk Extreme or Extreme Pro card. Peak bit rates in 24p mode can be 80+ Mbps, but the max average bit rate is around 66 Mbps.
@balazer - Do you know many people that have done quite heavy color grading of 24p shot with Cake? Does it seem to handle it well? I will probably use Magic Bullet Looks.
I look forward to your reply.
-Andy
I don't know. I try to do as little grading as possible. Please try and let us know how it works.
The club shots were with Driftwood's Moon, but the city shots from the monorail were with Cake 2.3. I darkened it a bit and color graded to match the club footage, but I was blown away with how nice this patch looks for night footage like this. I will try to post one of the original files soon.
WIll the manual movie mode(AVCHD1080I60) work well with this?
1080i uses the same encoder settings as HBR mode, so in theory they work similarly. But I've done no testing specifically for 1080i. Use at your own risk.
It looks like you're new here. If you want to get involved, click one of these buttons!