Tagged with gop15 - Personal View Talks https://personal-view.com/talks/discussions/tagged/gop15/feed.rss Sun, 28 Apr 24 19:15:20 +0000 Tagged with gop15 - Personal View Talks en-CA Driftwood Cluster X series 3:│moon T7│ЅріzZ T6│Nebula T7│Drewnet T9│ https://personal-view.com/talks/discussion/7580/driftwood-cluster-x-series-3moon-t7%D0%85r%D1%96zz-t6nebula-t7drewnet-t9 Mon, 22 Jul 2013 17:27:25 +0000 driftwood 7580@/talks/discussions A 'cluster' of different GOPs (Group of Pictures) forms the basis of the Cluster X series - to boldly see how far we can push each GOP factor in terms of superb looking artifact free Intra & prediction based Longer GOP (> 3) out of the panasonic cameras - where a natural scale of compression and consumed bitrate begins to appear according to its GOP size.

What's so good about all these new Cluster X settings? In simple terms, after looking at the rate control, quantisation and scaling matrices found in the GH3, Nikon D800 and Canon 5DMKIII I decided upon a radical approach to find out if the GH2 was capable of getting near their very low QP values (which range from predominantly 1 to 30 depending on the detail and luminance of the subject).

What I found was (and this had me concerned before) was that Panasonic's implementation / or ptools labelling!) of rate control using the GH2's quantisation parameter isn't too clear. I needed a way of having a minumum Quant and a desired maximum quant. Strangely with Initial Quantizer = 30, Quantizer for 1080 modes = 3 (+-2 can offer a QP of 1 !) and Quantizer for 720 modes = 30 I could sortof achieve this!!!!

With 720 using B frames (switched on in the 720 encoder 1 settings as = 3) Quantizer for 720 modes seems to work as a highest Quantisation setting (high is worse, low quant is best) together with Initial Quant of 30 (don't worry about such a high starting value here - after the first frame it settles straight back to lowest achievable (best) for the scene. Subsequently, Quantizer for 1080 modes = 3 gives me my lowest desired (best quant) for each opt 24p, 1080i, HBR and 720 modes.

This quantizer rate control now seems to work as in newer High Profile Level 5 (5.1) cameras such as the 5DMKIII and Nikon D800 (and seemingly the GH3) and when matched with a new softer looking matrix - as analysed on the Canon 5DMKIII (with adjustments) - brings a degree of softening to the over sharpe Pany lenses.

I'm also sure that 1080i/FSH?HBR modes work better with half the coded picture buffer size of 720p modes (indeed I switched off 720 Bottom setting (the buffer amount as in automatically works it out) whilst offering a fairly low buffer in bitrate settings in 'patches for testers'.

In summary, the range of QP will be (certianly in 1080p24 modes) from 1 to 30 - often staying very low (less than 10) all round. I believe (although only tested by myself) that the result offers extremely good looking 720p modes (probably the best ever), certainly superior 24p with Intra & Long GOP, and great looking HBR/FSH and iA mode control (though further strenuous tests may well be required). High dynamic functions work it seems too.

All the new Cluster X series settings for each GOP attached below have been tested on the notoriously sharp Pany OIS lenses. Give me feedback.

Its as simple as A, B, C, D, E, F ! Each .ini setting in ptools for Cluster X Series will equate to setA ('DREWnet' 12/15 GOP), setB ('Nebula' 6 GOP), setC ('Spizz' 3 GOP), setD ('Moon' 1 GOP/3 GOP). Any updates will stay on these sets.

"Cluster X Series isn't for pussies!" Driftwood 2013

If you enjoy these settings please don't forget to donate to personal-view by clicking on the link above. Gratitude and thanks to Vitaliy Kiselev.

]]>