AN "Valkyrie", GOLGOP3-13, GOP3ZILLA, natural, Freeze-less Reliable All-round GOP3 Settings :-)

bkmcwd

Since my main PC has crashed now, the tests in PAL of GOP3ZILLA-MAX_Standard_BETA1 and 2 are not made. :-(

It may somewhat be too aggressive.

In that case, please change some GOP tables as follows.

1080i50 GOP Table=1, 6, 6, 1, 14, 14

bkmcwd

Thanks for your feedbacks! :-) About 720p60, I am just going to consider which is the more nearly better between 6GOP or 9GOP. I refer also to the setting which you combined. BTW, how about 24p of GOP3ZILLA-MAX_Standard_V2_BETA1 or BETA2?

@kovtun66

"is there any documentation on what each number in the GOP Table means?"

I do not know about formal explanation, either. However, I understand the near meaning. Although this is my imagination to the last, I think that each parameter has determined the distribution ratio of the resource to various frames. Although I think that the P frame-oriented parameter is satisfactory at "0" in 24p of 3GOP, I also consider the parameter for the P frames by necessity in FSH and HBR of 3GOP. Although it will be fall back mode with the settings originally set to Q13 or Q15 under high stress, I discovered that the frame buffer was controllable by changing this GOP tables. This is judgment by my experience and I do not understand about the exact meaning, either.

karl

In the Flow Motion 2.0 thread I just posted an amalgamation of FM 2.0 with the 720p60 settings from GOLGOP3-13_standard, which solves the "involuntary stopping" issue for me: http://www.personal-view.com/talks/discussion/comment/65357#Comment_65357

I should also mention that I first tried to blend in another setting, Canis Majoris, to have at least one GOP1 setting in as an alternative, used with HBR (sacrificing all 25p/50p usuability). It worked, but the GOP1 recording was just too unstable, and it not only stopped at times, it froze the camera - and power cycling between shots is not an option for me when under water.

kovtun66

@bkmcwd: since you're one of the Masters of the GH2 world, I think I will ask you: is there any documentation on what each number in the GOP Table means? I see six numbers in every GOP Table and have no idea what each of them means. You explained about one of them some time ago, but what about the rest of? Thank you in advance sir.

karl

@bkmcwd: To answer your question from the FM2.0 thread: The GOLGOP3 version I was referring to when writing that it had never failed me in 720p60 mode was GOLGOP3-13_V4.3_BETA3 - which used GOP 9 for 720p.

Right now I am doing some experiments, trying to amalgamate FM2.0 for 24H/L, GOLGOP3-13_standard for 720p60, and Canis Majoris for HBR... I am not quite sure this will work, though... ;-)

bkmcwd

I release "GOP3ZILLA-MAX_Standard_V2_BETA1&2" settings. These are the experimental settings. In BETA2, the experimental "Debloking tables" are also included. These two are all the same except Debloking tables.

*About PAL mode, it is not tested enough yet.

Since my main PC has crashed now, the works do not progress. :-(

Although the setting which combined Sedna with natural is also made to the middle, the satisfying result has not been obtained yet...

Please download the raw data from vimeo site. :-)

bkmcwd

GOP3ZILLA_standard_V2_BETA4 HBR(30p) Motion Test

This is the experimental setting in HBR and FSH. The experimental setting of SH(60p) is also included.

bkmcwd

@towi

I understand well about the target which you aim at. It is because it is the same target as me. I am also going to inquire further. We are aiming for the same goal. Good luck to us. :-)

@kovtun66

I am improving your set. I am going to shoot in the forest after it. :-)

@sakattaq76

Thanks for testing my setting! :-) Looks really good.

sakattaq76

@subco, wait until the results of the "Revenge of the Great Camera Shootout" revealed. You will be convinced for sure, and buy three GH2s not just one. In my case I have tried many video cameras with cost up to 6000 euros and GH2 outperforms all of them. Annoying not to have some inputs-outputs of other "pro" cameras but you get used to it! Thanks to this forum and Vitaliy, made my mind and bought the GH2, never will regret that.

sakattaq76

Testing the GOP3ZILLA standard V2 BETA4. Different file sizes encoded with the best so far (IMHO) software "MainConsept Reference Encoder" from the original file (about variable 85Mbps and 900 MB). Shutter speed 100, variable ND filter. No color correction. Lenses used, Lumix 14-45mm, Canon FD 50mm f/1.4, Olympus 40-140mm f/4-5.6. Just press the "Download" button.

1280x720 24p @ 6Mbps (45ΜΒ) https://www.dropbox.com/s/ebm2hs3nb7j5nsw/GH2__Paraliakos%20720_24p_6Mbps.mp4

1280x720 24p @ 20Mbps (140MB) https://www.dropbox.com/s/zwpi4uzb1gbqbg5/GH2__Paraliakos%20720_24p_20Mbps.mp4

Full HD 1920x1080 24p @ 42Mbps (337MB) https://www.dropbox.com/s/azf27lfwmbtq7qt/GH2__Paraliakos%201080%2024p%2042Mbps.mp4

kovtun66

@Stray: thanks a lot for explanation. My mistake was, I thought it's just VBR approach in action. The recorded frames of the moving chart obviously had some motion blur, so I thought it's natural that less bitrate is needed to encode those frames...

towi

@bkmcwd

Honestly I don't have to try the latest settting. All in all I know it won't work as I've also tried to reduce P-Frame size with GOP-Tables. I am pretty sure the key for good HBR in 3GOP is the P-Frame scaling table... this is why I use the B-Frame scaling table for P-Frames in my latest 3GOP setting ( http://www.personal-view.com/talks/discussion/comment/63552#Comment_63552 ) and why I've posted promising HBR/25p settings for your patches above ( http://www.personal-view.com/talks/discussion/comment/63549#Comment_63549 ) . I would love to know how to design a Matrix (Scaling Tables) but I do not ... with the B-Frame scaling on P-Frames my settings are certainly not perfect yet, but it's getting closer to what I think should be a good HBR setting (fine grain, sharp static scenes, smooth motion ... well, we all want exactly that, I assume).

Don't get me wrong!!! I love your settings for 24H and I admire your really great work on these brilliant patches. However I feel there is much more room for improvement for HBR/FSH modes. For instance Sanity V3 or my latest modfied version of Sanity -> http://www.personal-view.com/talks/discussion/comment/63801#Comment_63801 (or especially 24p25-NINE: http://www.personal-view.com/talks/discussion/comment/45642#Comment_45642 ) show what can be done with HBR. Unfortunately these patches are long GOP and motion may look a bit harsh under certain circumstances (especially in bright, contrasty daylight scenes)... however for static scenes / tableau style work they look extremely good. My target is to match this IQ but with smoother motion ... but as I don't know much about encoder settings it's hard for me to get there (and a lot of detective work....).

Best Regards!

Stray

@kovtun66 No, it is the fallback mode. One of the reasons to do a static chart test is to check if this behaviour occurs. The fact that it happens at the same moment kind of proves it. Basically the camera cannot continue to encode at the level it currently is, and so it falls back to a lower level (compresses a lot more). Its actually less likely to happen in a real world shoot unless there is a lot of detail in the shot. The correct result from a chart test should be a continuous constant pattern of I-Frame and B-frame sizes with very little variation.

kovtun66

Actually, this was a static chart, and then I started the animation, and the drop occured at the same moment precisely... So, fortunately, this is not the Fallback mode... Anyways, thanks so much for your response!

LPowell

@kovtun66 The Stream Parser report you uploaded here shows an abrupt drop in I-frame size starting around frame 245:

http://www.personal-view.com./talks/uploads/FileUpload/54/c25e11f87fa7d418706c719074c64b.png

This indicates that the encoder was unable to sustain its initial high bitrate, and was forced down into Fallback mode, which uses low-quality quantization tables. Settings that exhibit this kind of behavior cannot be relied on in demanding situations.

kovtun66

@LPowell: oh... sorry, could you please explain how you know this from the stream parser? For example, here is the static death chart (by the way: in the REAL footage DC is most usually 5-7, QP 18-22 which, I suppose, is overall fine):

bkmcwd

@KevinTraywick

Thanks and please try my settings, supposing you are interested. :-)

@towi

Now, how is this? I was conscious so that FSH could also be used. :-)

This is the experimental setting in HBR and FSH. The experimental setting of SH(60p) is also included.

KevinTraywick

I haven't even started sorting though all of bkmcwd's work and this thread but I just made a donation based on his dedication to VK. This is great work and a great community! Drop a buck or two to keep it real if you can. Kevin

bkmcwd

@kovtun66

I am sorry that it becomes impossible for me to shoot in the forest since there will be my aunt's funeral tomorrow. And it is a rainy day today. :-(

Improvement in HBR mode is still under continuation. :-)

@karl

I am sorry that explanation in my English is insufficient.

"I am afraid I currently have no tool at hand which would allow me to find out on this easily."

How was qp in the result of JM-SP Decoder? If the value and range are known, I can do imagination about.

"Do I understand your statement "I think that GOP3ZILLA series realizes high IQ with the most sufficient balance" correctly in that you would expect GOP3ZILLA to deliver an overall better quality in the "moving leaves" scenario? Even though GOLGOP3 was described as using a higher average bandwidth?"

Yes. The difference between GOLGOP3-13 and GOP3ZILLA is a difference in an elated scene. On the high stress scene, like "moving leaves", GOP3ZILLA may become high IQ rather than GOLGOP3-13 as a result.

@hil130

"Which version should we use in GOP3ZILLA standard V2? BETA2 or BETA3?"

Since these are all the experimental settings in HBR mode, use depending on conditions is possible for them. However, I understand that it is necessary to improve BETA3 more. Of course, I am working on now. :-)

LPowell

@kovtun66 Sorry, but the Stream Parser report shows the encoder plunging into low-quality Fallback Mode on the left side of the frame chart. This indicates that the camera was not able to sustain the initial high bitrate and was forced to degrade the image quality to avoid crashing.

hi1130

@bkmcwd: Which version should we use in GOP3ZILLA standard V2? BETA2 or BETA3? Thanks for your awesome work!!!

karl

@bkmcwd: Regarding "How was qp of the macroblocks of this portion?" - I am afraid I currently have no tool at hand which would allow me to find out on this easily.

Do I understand your statement "I think that GOP3ZILLA series realizes high IQ with the most sufficient balance" correctly in that you would expect GOP3ZILLA to deliver an overall better quality in the "moving leaves" scenario? Even though GOLGOP3 was described as using a higher average bandwidth?

kovtun66

@bkmcwd: thank you... would be very cool! Just by the way: it SEEMS like to get HBR closer to 24P the B-frames shall be as large as possible. This is because of the way interlaced encoder mode works.

bkmcwd

@kovtun66

Thanks mate! :-)

"if you get time and inspiration, could you please try to film a traditional fly-around in your small forest with this setting in 24H?"

Since tomorrow's weather report rains, I will try, if it is fine the day after tomorrow.

@towi

It seems that BETA3 is still not good. :-( I am improving succeedingly.

kovtun66

Pals, I have filmed a lot in 24H this weekend using a following merged setting and I was very satisfied. Basically, it's my merge of bkmcwd's GOP3ZILLA with SednaAQ1C, MAX frame/FB/SB limits and 'natural' Q18. GOP table tweaked a little bit, that's all from my side. It's obviously bkmcwd who takes all the credits. The result is really good, IMHO. Fat B-Frames/Sedna matrix cotribute to low-light performance, while Q18 and MAX I-frames make the setting very immune to infamous macroblock level lines artifacts. By the way I have seen 1160KB I-frames in the real-life footage I filmed with this setting. The beast? It could be, for those who don't care about a guaranteed in-camera playability of the 24H footage. @bkmcwd: if you get time and inspiration, could you please try to film a traditional fly-around in your small forest with this setting in 24H? Thanks a ton already. Cheers! P.S. Ignore other modes, it's intended as 24H-only test setting...

Howdy, Stranger!

Categories

Tags in Topic

Top Posters