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

bkmcwd

Thanks for the info! :-) There is no telling whether to actually adopt to my setting, yet. Although I have not seen the newest matrix of @driftwood in detail yet, adjustment is somewhat required to combine.

bkmcwd

@soru

"I also feel the noises of Sedna resembles to intra's."

I also feel so. :-) Processing of the noises is somewhat difficult for me.

bkmcwd

@har_k

Thanks, mate! :-)

Although GOLGOP3-13 is not the setting which targeted span positively, I tweak it a little.

ishvar

@bkmwcd Did a testshot comparing your Natural- and C.M. soft skin of Nick Driftwood. C.M. recorded 2.3X higher in bitrate than Natural. C.M. gives a cleaner image with smoother curved details and less greens in the blacks but Natural is a great medium mbs patch! Coming days i will also compare a few higher bitrate GOP 3 patches of yours with GOP1 C.M. Thanks as always! Vimeo video: C.M. @ the left and Natural @ the right

picture caniscroptest: upper prt. is natural - lower prt. is the shiny dog with soft skin - lol

edit: the name of the last picture should be natural not golgop!

Ramses

GOLGOP and Canis Majoris don't mix ;-)

soru

@bkmcwd

Thanks.

While I don't do check exactly with side-by-side test, I also feel the noises of Sedna resembles to intra's.

har_k

@bkmcwd I also read pm. And, I sent pm.:-) Thanks.

bkmcwd

@har_k

Many thanks for feedbacks always, mate! :-)

If positive span is required in FSH, is it necessary to tweak the setting more?

bkmcwd

@soru

"Sedna has bigger QP & B-frame size than the other."

Yes, this is the feature of the Sedna version as a result. Since according to the guess which I heard from @Stray it subdivides more finely and the codec is processing the picture, the B frame is whether do become large and qp also becomes high. Although I do not understand truth well, the grain noises will be felt finer if I actually look at the picture.

"Bigger B frame size in the motion like your floating sample seems to be good, but generally Lower QP is desirble."

I do not think that the B frame of GOP3ZILLA_StockMatrix is small. Even if I look at other samples, sufficient size required for the scene is secured. I do not say that a difference comes out in motion, but consider that the difference has only appeared in the size of the B frame as a result of the difference in the processing described above.

"Which is better? Is it the choice of kind of the user's liking?"

Of course, it is based on your liking. Although this is based on my subjectivity to the last, is GOP3ZILLA_StockMatrix the highest if it is the purpose to obtain the highest feeling of resolving also in what kind of scene?

On the other hand, probably, the Sedna version of GOLGOP3-13 or GOP3ZILLA is suitable for those who want to use a Sedna matrix more effectively with bigger frame size than Intra settings.

har_k

@bkmcwd, Thanks for your new settings!

Since the weather is not good, I tried long time recording in the room. The subject is my dirty jeans, and SD card is SANDISK Extreme Pro SDXC 64GB.

HBR spanned 4 times and stopped (4GB+4GB+4GB+4GB+4GB, recording time is about 43 minutes, bitrate about 65Mbps)

FSH spanned 1 time and stopped (4GB+4GB, recording time is about 17 minutes, bitrate about 66Mbps)

SH spanned 4 times and I stopped manually (4GB+4GB+4GB+4GB+3.1GB, recording time is 60 minutes, bitrate about 44Mbps)

*"GOLGOP3-13_V4.14_BETA3", iso3200, F5.6, SS1/60, NOS(-2,0,0-2)

soru

@bkmcwd I saw the attached report of streamparser.

So, the differences between the settings with and without Sedna are QP value and B-frame size? Sedna has bigger QP & B-frame size than the other.

What does it mean?

Bigger B frame size in the motion like your floating sample seems to be good, but generally Lower QP is desirble.

Which is better? Is it the choice of kind of the user's liking?

bkmcwd

@liquidify

Thanks! It is very interesting. Please give me feedback also to me. It is compared with the great setting of Nick after a long time. :-)

liquidify

now to test your new golgop3 with sedna vs canis majoris!!

bkmcwd

I release "GOLGOP3-13_V4.14_BETA3", "GOLGOP3-13_V4.3_BETA4", "GOLGOP3-13_SednaAQ1C_V2.1_BETA4", "GOP3ZILLA_V2.13_BETA6" and "GOP3ZILLA_SednaAQ1C_V2.1_BETA2" settings.

In these, some of the GOP tables are same experimentally. These are the almostly same settings except Q value and matrix.

Special thanks to @driftwood! :-)

Please try them!

Personally, I prefer "GOP3ZILLA_V2.13_BETA6", "GOP3ZILLA_SednaAQ1C_V2.1_BETA2" and "GOLGOP3-13_SednaAQ1C_V2.1_BETA4". It is because the codec does not waver in any scenes and the level lines of the high qp macroblocks do not appear with those settings.

Especially "GOLGOP3-13_V4.14_BETA3" is suitable for lowlight.

Although the codec with "GOLGOP3-13_V4.3_BETA4" cannot waver easily, behavior unlike GOLGOP3-13 series is carried out except under the lowlight.

*Already combined with @thepalalias's "Pasadena_Pulse_Audio_V2_Beta2"

NOTE:

*Although bit rate setups are quite high, the bit rate which actually appears are lower than these. Since the actual bit rate changes by the object to shoot, please confirm by yourself.

*24H and 24L are high quality 3GOP setting with mega size I-frame.

*In GOLGOP3-13_SednaAQ1C, 24p80%(30p) is available in only 24L.

*In camera playback is available in 24p. In other mode, Sometimes available after reboot the camera.(Not tested yet, but probably)

*EX Tele works.(Not tested yet, but probably)

*Spanning in 24H, 24L and SH with SanDisk 95MB/s 64GB SDXC Card. In other mode, need test.

*Spanning in other cards needs to be tested.

proaudio4

@ishvar

Wow! Really great job. Amazing. very clean

bkmcwd

@karl

Although I am not detailed at all about ffmpeg, I am very interesting. :-)

karl

Good question. I've taken a look into the source code of ffmpeg, and it seems that the maximum valid sizes for cbp do not depend on any encoding parameter, the limit is coded in the source, see:

http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavcodec/h264_cavlc.c;h=a4debade2e01cabc197171613a3379641179a2ab;hb=HEAD#l1060

The comparison has been in there for almost a decade, the "cbp > 47" check was introduced into the code in 2003 here:

http://git.videolan.org/?p=ffmpeg.git;a=blob;f=libavcodec/h264.c;h=757400b3372243c3e0eeb23dae5d8317ca42c235;hb=0da71265d84b587c7159cd82708ca60ad050dd4c#l3511

Alas, my knowledge of H.264 is not detailed enough to know whether this limit check is correct, but given the many regression tests ffmpeg has gone through, I would be surprised if this was a 10 year old bug in ffmpeg...

bkmcwd

@karl

Thanks for detailed report! :-) I also often look at "glitch" in MediaPlayer, especially in HBR. In VLC, not seeing is also the same of you. If it is such a thing, it is not a problem of the setting itself?

karl

@bkmcwd: It's good that you asked, because now that I took a more detailed look at the 720p60 clip with the "glitch" I noticed that this glitch is not exhibited by every player software - VLC and kdenlive do not show it, while mplayer does. Mplayer emits the following message when it comes to the point of the video where the glitch is:

[h264 @ 0xfd6920]cbp too large (57) at 73 34 [h264 @ 0xfd6920]error while decoding MB 73 34 [h264 @ 0xfd6920]concealing 856 DC, 856 AC, 856 MV errors

("cbp" here means "coded block pattern", as explained in e.g. https://docs.google.com/viewer?a=v&q=cache:yf_vIxYGiRgJ:www.cc.ntut.edu.tw/~shyang/Journal%2520Papers/H.264%2520Fast%2520Inter-Mode%2520Selection%2520Based%2520on%2520Coded%2520Block%2520Patterns%2520_final_.pdf+h.264+cbp&hl=de&gl=de&pid=bl&srcid=ADGEESil4BsBzTs_LZocsq2gJXFiAXbqzLgY6QMsInTp4ScLH1TZgDVJo2HXHzrKnyBwVVzl2_up6pb1Zfnq-lGAK8Fj3R5861FR6H82LtsR4mgUgUJ-0RY5wP7TwxVL4Wa9kFe8-Yo2&sig=AHIEtbS6ib2kncHptoXXuWxIUoEyajcX2Q )

Since VLC and kdenlive under the hood use the same ffmpeg library that mplayer uses to decode the H.264 stream, I assume they just a more bug-tolerant parameter setting while using ffmpeg.

The glitch is also visible when using hardware-accelerated decoding (using vdpau on a NVidia GFX hardware).

bkmcwd

@soru

Thanks always, mate! :-)

60p is under improvement now in 9GOP. Please also try the following high version.

bkmcwd

@Ramses

Thanks for testingmy setting in GF2! :-) Since I do not have GF2, I have not tried by myself. I would also like to look at the setting which you combined.

bkmcwd

@karl

Thanks for testing my setting and feedback! :-)

"720p60(SH) also looked good (rate was ~ 40MBit/s), but there was a glitch where 3 consecutive frames contained some garbage in the lower third of the screen."

Would you explain with a screen grab what kind of problem there was concretely?

"HBR was recording at a rate much lower than 24H - is that intenional?"

In 30p or 60i, it turns out that the bit rate goes up neither 25p nor 50i at least with 3GOP. Of course, naturally in this setting, the bit rate does not go up on a par with 24H.

The reason for having set the pretext as 80M is stabilized if this value is made higher than the real bit rate.

ishvar

@soru Beautifull slo-mo material! Keepup the good work!

soru

Hi. I tried GOLGOP3-13 SednaAQ1C v2.1 on 720p and made a slow motion footage.

all movies have around 20~40Mbps. first, the water scenes have rather lower bitrates (around 22) and higher QP like over 30. then next, the highly detailed nature scenes or geometric floor scenes have rather higher bitrates(around 30) and middle QP(20-27). the most interesting for me is that if there is a human ,and however small the subject size is in the resolution, then the QP of the frames becomes lower even in highly detailed scene.

maxmizer

@Ramses, you can put your modified pach for gf2 attached here

Howdy, Stranger!

Categories

Tags in Topic

Top Posters