GH2 Cake v2.3: reliability and spanning in 720p, HBR, 24p, and VMM at 2-2.5x stock bit rates

arvidtp

i shot some interior shots today with cake 2.3b1 and they averaged only 38 to 41 Mbps on 24H. So i re-flashed with cake 2.2 and sure enough i was getting 61 Mbps again with the same scene. I feel like the quality was a little better on the cake 2.2 shots than the 2.3 shots (makes sense because 1.5x the bitrate) but it was pretty subjective and had trouble picking out any one thing that i thought was worse on those shots that i could screen capture and show here, especially since they were hand held.

balazer

Ugh. I can't give you a higher bit rate and have it never go into fallback mode, while also making it span all of the time. Pick two. Everyone was perfectly happy with the quality, and then someone had to go looking for problems in Stream Parser.

arvidtp

@balazer don't worry! if that is expected behavior, its fine - just thought it might have been out of the ordinary to have a lower bitrate :) I actually like having a little lower bitrate myself, since i often shoot performances - gives more time on the card. And as i said - was very hard to pick out any real quality difference. Thank you once again for an awesome hack setting.

balazer

arvidtp, I think I got it now. Please try this. These settings are just like Cake 2.2, except I've changed the fallback mode scaling matrices to be the same as the B-frame scaling matrices, same as I had done in previous versions of Cake. Now I'm seeing a nice pattern of regular frame sizes in fallback mode, and great quality, just above 60 Mbps in 24H. It's stable and spanning in 24H. If you want a slightly lower bit rate, use 24L.

Edit: sorry, 2.3b3, not 2.3b2

mrttt

So are we all meant to use beta3 now and report our findings? Is beta3 now the "best" cake to use?

balazer

I'd like people to try 2.3b3 and compare it to 2.0-2.2 in 24H. (2.0-2.2 are all the same in 24H)

luxis

@balazer thanks for the cake! great taste ;) i used cake 2.2 and was quite happy to see the great results with modest file sizes, thank you for your hard work on it! will load and go test 2.3b3 now.

arvidtp

back up to ~58.5 Mbps in 24H with 2.3b3 here. Quality seems similar subjectively to 2.2.

However, as long as 2.3b1 proves to be as stable as / spans as well as 2.2 (do you have any reason to believe it would not be?) it may very well be my setting of choice for shooting concerts since i calculate ~1:50 hours on a 32GB card using 24H (assuming 40Mbps), and ~2:08 with 24L (assuming 35 Mbps), which is just about a perfect quality/record time tradeoff for the concerts i tend to shoot, especially since the camera tends to sit on a tripod and is seldom moved, hence lower bitrate is usually OK.

I can see 2.3b1 being relabeled as "cake lite", or maybe "cake, low fat" :) and posted along with standard cake, as long as it is as stable as the other versions.

balazer

Thanks for the testing and feedback, everyone.

arvidtp, I think 2.3b1 should be stable. I set it the same way I did in all of the other modes, putting the max sustained bit rate low enough to never go into fallback mode. But the bit rate in 24P is too variable, making the typical bit rate much lower than I wanted or expected. I'm guessing the variability is because of the short GOP length, which is shorter than I set in all of the other modes. (The GOP length in HBR is actually 6, not 3, because the GH2's interlaced coding uses field pictures) Instead of using 2.3b1, I'd suggest using 2.3b3 in 24L mode. If that bit rate is still too high for you, you can try lowering the 24L bit rate 10 Mbps or so. But if you change anything, you're on your own. Be sure to test your changes.

Short GOPs are less efficient than long GOPs, so they're not suited to low bit rates. The only reason to use short GOPs in the GH2 is to support spanning at higher bit rates. But since short GOPs are less efficient, if you use a short GOP with a low bit rate, the quality is lower. If you really need lower bit rates, I'd suggest using other settings like Cbrandin 44M or Sanity.

BillTheButcher · Screen shot 2012-04-20 at 6.29.17 PM.png

Finally got a chance to test out Cake 2.2. So far VERY pleased with the results.

adventsam

Does cake 2.3b1 have b frames in 720p? Only I am sick and tired of 1080 with all that stutteryness, have tested extensively and I can truly see that much at normal viewing between 1080 and 720, only when I have my nose against the screen and the most demanding stuff. The 720p/50 is buttery smooth, can be thrown around and to cap it all the af works so effectively too.

mk47

Just tried Cake 2.3b3. I only tried a quick timelapse to test spanning. It worked fine but the final bitrate of the file was around 28mbit. Im guessing that was caused by the nature of the scene with its low contrasts and still scene. Remaining time was still being show for 70 mbit.

balazer

Adventsam, Cake does not use B-frames.

smontoto

@balazer I ran a test on 2.3 b3 with both 24h & 24L modes. They both spanned fine the entire card. I have a question though - On 24L the first clip played fine in Windows media player but the rest of the clips only showed black screen and no sound. When I put it on my editor it plays fine. 24H did not have this issue.

24H- Avg 64m/s 24L- Avg 49m/s

Both were on a static detailed scene.

balazer

I can't say anything about playback in Windows Media Player without knowing what DirectShow filters were in use. I use Media Player Classic Home Cinema, and its built-in filters work fine, as does ffdshow-tryouts.

smontoto

Downloaded Classic Home Cinema and it and 24L clips play fine. Something weird with Microsoft Media Player.

The clips look great, I cant tell any difference really from the 24h. Maybe in low light/high ISO?

Thanks again!

mrttt

this is my cake 2.3b3 test

luxis

i see a little issues with the dark yellow boxes around 1"38 ...is this the same in the original file or youtube artifacts? oh and i see later cake filming cakes ; )

mrttt

I will investigate. I did notice a possible auto white balance issue around 1 minute. I'll need to remember not to use auto. Do you mean on the brown boxes as the kid steps away from them?

*It does look like there's some dithering on the boxes in the original file

luxis

yep right there

frerichs

Used Cake v2.2 and it worked flawlessly. I let the interview roll for over an hour with no issues spanning and in post I did literally next to nothing to match the GH2 and AF100 footage. Truly a piece of cake!

For the GH2s, Voigtlander 25mm f/.095 and LOMO OKC1-50-1 2/50mm lenses using 24H, Vibrant -2 -2 -1 -2, ISO 200 and K3200 for white balance.

The AF100, using Olympus 14-35 f/2.0 lens and a profile to match the GH2s as following (ISO 200, K3200):

1080/24P Detail Level -4 V Detail Level -1 Detail Corning -4 Chroma Level +2 Chroma Phase -3 Master Ped -5 DRS Off Gama Cine-Like D Matrix Fluo

Edited with FinalCut Pro7

The GH2 camera angles and her right side wasn't lit well, but it's a no budget on the fly production ;-)

kazuo

hey @frerichs moving testimony, but why is her voice in mono?

feha

@balazer , I'm now testing Cake 2.3b, however I'm just curious why B‑frames are left out ? I'm not an expert in codecs, does this have any impact in motion, panning etc ? I like Cake :-)

balazer

In not using B-frames, I lose some efficiency, but gain frame-to-frame consistency. Also, it's a recipe that's proven stable in all of the different modes, particularly in HBR.

gameb

I am shooting with Cake 2.3b3 24L since a few days, doc stuff, everything looks great, very happy, thanks balazer

Howdy, Stranger!

Categories

Tags in Topic

Top Posters