@bkmcwd what did you change in this current version?

@Tobsen @sebasp1

"what did you change in this current version?"

Stability and image quality have been improved by change of GOP tables. :-)

Since there is a tendency for qp to become high rather than using stock matrix when Sedna matrix is used, I think that I optimized how depending on which qp appears in 24p.

Although this was made in 3.65d, of course, it could use also in 3.64d.

BTW, since I have improved 720p now, I will release some new settings soon again.

I loaded it up this morning (12 hours ago) and shot some daylight clips downtown and around the harbor. GOLGOP3-13_SednaAQ1C_V2.* just keeps getting better. Shot some 1 minute 'stairwell torture' clips. File size for 30p (around 311MB per minute) was a bit smaller than I expected (I'm a bandwidth hog) but it's smoother in low light (ISO 3200) than beta 1. 24p also looks 'smoother' (and cleaner) than beta 1. I had no problems with write errors or in-camera playback shooting 1 minute clips with Cinema 24p. Didn't check spanning. Thanks for doin' what ya do! much appreciate it :-)

@bkmcwd, wow GOLGOP3-13_SednaAQ1C_V2.1_BETA2" is amazing. It's the best yet from you. I tried 24H and it looks, wow. Great work man!

@HillTop1

Thanks! :-)

@liquidify You can't really change the Q (AQ being a different approach to the same thing) as the Q settings are probably specific for the GOP 3 setting. What @bkmcwd has done is to incorporate the matrix from sedna C and then rebalanced everything for it to work effectively with a GOP of 3 (which probably has involved setting the Q to a different value). So setting Q20 or AQ1 is meaningless in this particular GOP3/Sedna C scenario per se. My understanding of @driftwood s settings are that the matrix of the Q20 and AQ1 versions are identical, and its just the quantizer that is being set differently (AQ1 kind of forcing a higher bitrate accross more shooting scenarios).

I think what @bkmcwd means by puzzles the codec is that it is not entirely stable, what I noticed myself was that the B-Frames would be extraordinarily large even on static shots with the A matrix. There are also versions of the GOP3 settings with the Sedna A matrix in this thread if you want to try them.

@bkmcwd I was interested in testing your latest settings to evaluate how well the Sedna Scaling Tables work with B-frames in short-GOP sequences. I used a variety of well-lit subjects - high-detail, fast-motion, and low-contrast. The videos recorded reliably and look good, but the statistics I saw in Stream Parser lead me to suspect that Sedna is not well-matched to these settings. It consistently produced QP levels in the high 20's and DC quantizers of 7-8 in all video modes. The only case where I was able to produce DC quantizers of 5-6 was in 24H mode shooting a low-contrast, static subject.

Looking over the INI settings file, I see that Q is set to 13 for all video modes, which is a far more aggressive global quantizer than the encoder was actually able to use, in spite of 24H mode's ability to produce very large I-frames over 1Mbyte in size. My hunch is that the very small quantizers used to encode fine details in the Sedna Scaling Tables are overkill for this patch, and use up so much of the available bitrate that the encoder is unable to achieve your intended QP values and DC quantizers.

@liquidify

What Stray described is almost the same as my understanding. :-)

@Stray

Thanks always, mate. :-)

@towi

I have also tried the same matrix before. I was also able to obtain the good result. However, since there was an opinion whether the matrix for the P frames should also be used, due to interlace processing then at FSH or HBR, I stopped using the matrix. Though 1080p24 is good with it, how do you think in 1080i?

@bkmcwd With the Quantizer level manually set to 13 in your patch, the encoder will initially attempt to use 13 as the QP value for each frame. If that level of quantization requires too much bitrate, the encoder will try higher QP values until it produces a low enough bitrate. As we can see in your Stream Parser screenshot, the encoder is actually using QP values that range from 19-21 (in 24H mode). Combined with the Sedna Scaling Tables, that produces DC quantizers in the 5-6 range. With the GH2's max frame size limits, I think this is probably optimal, since as cbrandin points out, DC quantizers below 4 are counterproductive anyway.

This would all be fine if the patch produced DC quantizers in the 5-6 range in all video modes with a wide range of subject matter. Unfortunately, my tests showed that it could only do this in 24H mode with static subjects. When shooting in other modes, QP typically rose to around 25, producing DC quantizers in the 7-8 range, a few steps short of optimal performance.

If you examine the Sedna Scaling Tables, you'll find the intra quantizers range from 5-10 (except for the finest details, which are completely suppressed). This is very flat scaling and will require considerably more bitrate than the encoder's default scaling tables. That's why I suspect the Sedna tables aren't well suited to the low QP levels you designed the patch to work with.

@bkmcwd just wanted to let you know that i have been testing your GOP3ZILLA SednaAQ1C Vers 2.1Beta and i am getting great results in day to day use. very good detial; no visible banding or blocks; good noise rendition; and all at reasonable bit rates.

i am no expert with Streamparser but it also gives me good results whatever i throw at it with this setting (high detail or high movement, etc.) QP range is consistent between 1-5, and DC always at 4 or 5.

many thanks for your great work.

Hi Bkmcwd. I'm a (too much) silent user of your patch for some week, very good work, thanks.

I like driftwood's Intra patch but since it don't drive very well for PAL version, i stick with yours.

In HBR PAL, GopZilla 2.12: good all around image, motion not so bad, average bitrate near 50Mb/s and good recording time. (1H = 32 Gb).

For me, it's the best HBR/FSH PAL all around Patch at the moment. Don't use 720 50p, so id'ont know..

I agree with Towi and Lpowell, i find Stock matrix version of Gopzilla and Golgop better than Sedna version.

I don't post so much so Big thanks too at VK, NDriftwood, LPowell, Cbrandin. Special Thanks to you Bkmcwd and Towi for your work on PAL version!

Thanks