@Zaven13 Yeah, can you give it a trial run? I'm still pondering...

@Mirrorkisser @abraham1307 @cjdincer

I am starting the test for improving 'Spizz' requested from Nick. I think that this takes time for a while. Many thanks for your feedbacks about Spizz. :-)

@bkmcwd here is a HBR sample from Cluster X 6 GOP Nebula: Best HBR footage from my cam so far.(just converted with compressor for youtube) Other modes are also ok. tonight I've loaded 'Spizz' and give it a try couple of days. excellent setting 6 GOP nebula is. I hope 3GOP is better.and very curious about ClusterX intra VK. Also have to mention intravenus V1 is my personal best. Thank you Driftwood.

all -2 smooth iso160

@RKM Bart, that looks very promising - and with EX Tele! :-) Excellent shot.

@driftwood: Spizz sounds super interesting! Trying it right away. For which scenarios did you invent the new matrix? Cheers and thanks for your efforts!

Wrong thread

Wrong thread

OK, I didn't knew you were using some objective measure to access IQ, that's a good systematic approach, even though PSNR/MSE is a very poor (although simple and very fast to compute) when compared to SSIM: https://ece.uwaterloo.ca/~z70wang/publications/SPM09.pdf

I was about to ask you how on Earth was StreamEye reporting PSNR values without a reference (uncompressed video) but after some research (and finally reading the manual) it seems that it's able to compute a very good estimation (thus the name E-PSNR) from the encoded stream (macroblocks, headers and quantisers). Well... it's better than nothing, and still very useful for settings development. :)

Time to do some patching...

Ok, I'll wait, thanks again! Meanwhile I'll test your advance.

Q16? Isn't it overkill? I've read reports saying that SSIM values of 0.985 are near transparent and my x264 exports regurlarly exceed that at Q20. I'm probably comparing apples and oranges because any similarities between the GH2 and x264 regarding h.264 profiles, might be a pure coincidence. I even seriously doubt that the GH2 codec is tuned for SSIM, probably it's just PSNR. No CABAC either, no dynamic DCT partitions, different scaling matrices, etc...

Taking these considerations I understand perfectly that your priorities are on absolute IQ, while mine are more focused on efficiency and "just" maintaining visual transparency, but since you have a lot of "Pro" settings, I'm grasping at some "diversity" here, in the lines of a Sanity sidekick. :)

Great work!

Cheers Nick @driftwwod.

First of all I'd like to thank you for working on long GOPs too.

The trials links might be more pinpoint like this: http://www.personal-view.com/talks/discussion/comment/105936#Comment_105936 and http://www.personal-view.com/talks/discussion/comment/106345#Comment_106345

For efficiency reasons I've very much interested in these settings. Which matrices are you using in them?

If you don't mind, it would be nice if you set a purpose/mission statement/small description in the first thread. :)

I've been stuck to Cluster V6 DrewNet as it seems to be the only one which spans forever. 6 GOP Nebula ran for 10min before stopping. Are any of the other X versions likely to span reliably?

Looking forward to this series.

3 GOP - Hi-Quality - Pro Motion 'Spizz'

Update News (03/03/2013): Spizz Trial 5 released based on some of the ideas from Nebula and DREWnet. Its a safe QP15 and seems to work on all modes including iA (Intelligent Auto) @bkmcwd and @towi will be testing these findings and improving them even further. More possible updates on Spizz with bkmcwd's and towi's findings soon.

HISTORY

Cluster X Series: 3 GOP 'Spizz' from Driftwood, bkmcwd and with the help of @towi Trial 5 (Released 03/03/2013)

Updated Matrix and NO frame limit to cater for the hi-bitrate superior looking files. 1080p24 now records at a minimum of around 90Mbps on low detail stuff and 137Mbps at high detail. 720 SH/1080iFSH now work on iA modes, HBR modes should be very high quality.

Many thanks to @bkmcwd and @towi for improving these settings.

Cluster X Series: 3 GOP 'Spizz' from bkmcwd Trial 4 (Released 12/02/2013)

An update and 'tune-up' to Trial 3 with stability in 720p modes. Matrix and frame limit amendments to cater for the hi-bitrate superior looking files.

Cluster X Series: 3 GOP 'Spizz' Trial 3 (Released 06/02/2013)

An update to Trial 2 with stability in 720p60 by utilising a 6 GOP. Also the 24p L setting is now fashioned to 72Mbps i/b frames for fast action and more ameniable quality than the previous 50Mbps L setting.

Note: 3 GOP 720p60 is still being worked on by bkmcwd amd 'may' appear in Trial 4. However, the 6 GOP as usual is the best there is.

Cluster X Series: 3 GOP 'Spizz' Trial 2 (Released 03/02/2013)

An update to Trial 1 with stability in HBR/FSH/720p modes and better overall constant picture QP.

Cluster X Series: 3 GOP 'Spizz' Trial 1

• All round 3 GOP from 720p, FSH 1080i, HBR, and top rate 24p all using the new balanced Cluster X series matrix.
• Very Hi-quality reliable setting including upto 117Mbps 24p.
• Unique 3 GOP 720p implementation with great QP upto 76Mbps Hi Scene detail / 40Mbps avg Lo scene detail.
• Hi quality HBR Progressive 25p/30p and FSH 1080i interlaced modes. Upto 80+Mbps Hi-detail scenes.
• No Frame limit on any of the 3 GOP settings and ALL utilising B frames.

Further tests & releases of these settings will be conducted by @bkmcwd and @towi.

Spizz in action:- 3 Pictures of 24p, HBR25p & 720p50 (Spizz handles 720p50 at 3 GOP - Flowmo uses 6 GOP) modes demonstrated against another popular 3 GOP setting. You be the judge.

24p

HBR25p

720p50