@apefos: You're the man!!!

While some others make a patch within 3+ months, you did something really great that no one did before, just in few hours!

The Patch - 12 versions (download link below):

all versions have all improvements from my previous developing in workhorse, boson, gspot, dct and apefos settings, plus the new achievements on correct matrices to get great noise/texture and deblocking tables to avoid the diagonal rain pulsing pattern.

best gop settings and data distribution among I, P and B.

best frame limit and buffer sizes

6 different matrices in 2 datarate options: stable and high = 12 versions

stable versions can use auto electronic features and native lenses

high versions does not work with auto electronic features and are more stable with manual lenses

only version G is fully tested to avoid the diagonal rain pulsing pattern, I consider it "the patch". also it does not delay the on screen info display.

other versions are not tested on the display info delay, see instructions to test them in the txt file.

other versions have same deblocking tables, but I did not test the rain issue on them, there are instructions in txt file inside the zip to test.

the deblocking tables uses value 18 for intraframe and value 6 for interframe (all positions), it solves the diagonal pattern pulsing problem in setg, so I repeated it in other versions.

other versions increases or decreases quantization to get better textures in other isos, see the txt.

I perceived that lower quantization is better for lower iso and higher quantization is better for higher iso, so there is 6 versions with different quantizations to chose.

the quantization matrices are (all carefully tweaked in frequency increase ratios):

sete, lowest quantization,.

Luma:

006 008 010 012

008 012 014 016

010 014 018 020

012 016 020 024

Chroma:

006 008 010 012

008 012 014 114

010 014 120 126

012 114 126 156

setf, lower quantization.

Luma:

006 009 012 015

009 015 018 021

012 018 024 027

015 021 027 036

Chroma:

006 009 012 015

009 015 018 114

012 018 120 126

015 114 126 156

setg, standard quantization, "the patch"

Luma:

006 009 012 018

009 015 021 027

012 021 030 036

018 027 036 054

Chroma:

006 009 012 018

009 015 021 114

012 021 120 126

018 114 126 156

seth, higher quantization

Luma:

006 009 012 018

009 015 021 036

012 021 045 054

018 036 054 072

Chroma:

006 009 012 018

009 015 021 114

012 021 120 126

018 114 126 156

seti, highest quantization.

Luma:

006 009 012 018

009 015 021 054

012 021 060 072

018 054 072 108

Chroma:

006 009 012 018

009 015 021 114

012 021 120 126

018 114 126 156

setj, extreme high quantization

Luma:

006 009 012 018

009 015 021 063

012 021 072 084

018 063 084 126

Chroma:

006 009 012 018

009 015 021 114

012 021 120 126

018 114 126 156

@apefos @producer Guru Apefos really decoded the matrix. I will soon try this new patch.

updated information in previous post about the versions

After the creative discharge, some sleep, food and rest, I did a look on the versions and I found some things that could be better:

in setf luma matrix, the last position is 036, I think 033 makes more sense to the matrix maths.

in seti luma matrix, the number 060 can be 063 for better maths in the matrix.

there is another option in deblocking tables to try: 18 and 3 instead of 18 and 6, needs tests.

it is possible to have another version between setf and setg with this matrix design:

006 009 012 018

009 015 021 024

012 021 027 030

018 024 030 036

I also created some ideas to implement fallback. the fallback versions can use this design: a lower quantization in main matrix and a little higher quantization in fallback matrix, this way when there is heavy motion or iso increase the fallback matrix takes place to avoid compression artifacts.

the idea is to combine versions:

main matrix E + fallback matrix F

main matrix E + fallback matrix G

main matrix F + fallback matrix G

main matrix F + fallback matrix H

main matrix G + fallback matrix H

main matrix G + fallback matrix I

only drawback can be the fallback takes place when you does not want, for example in low light steady shots, if this happens textures will not be so good as the patch without fallback, but if fallback is just a little different the result can be good a balance...

setg in datarate stable folder

Thanks for the work apefos.

Forgive my ignorance but can I combine setg,i, and j into one patch?

For example use camera settings 24H,24L, and HBR(Pal) to change between setg,seti, and setj? or can 24L not use 24H patch settings?

There is a way to combine patches using different matrices for 24p, hbr and 720p, because the firmware uses different matrices for each. This way the different modes can have different matrix design and quantization based on their needs.

But 24H and 24L uses same matrix. HBR, and 1080i FH and FSH uses same matrix, 720p SH and H uses same matrix.

The other way to combine patches is to use the fallback tables, using a little higher quantization in the fallback.

The fallback works like this:

the main matrix is doing the job to compress the image. if heavy motion occurs or if you increase the iso more than the main matrix can handle, then the fallback matrix takes place to avoid artifacts, so using a higher quantization in the fallback is the way to combine patches. see my previous post.

Both things can be done, a slightly increase in fallback quantization and different matrices for 24p, hbr and 720p.

updated versions with one more quantization matrix option and deblocking tables 18-3

please read the txt inside the zip.

the new matrix is:

006 009 012 018

009 015 021 024

012 021 027 030

018 024 030 036

from now I am thinking about the fallback implementation and about different matrices for the recording modes, if these two improvements worth for better quality and how to do best settings for them.

@frullaccia the C version is the most high quantization amount in the matrices. it can increase the mud in textures under normal conditions, but if it is useful in these extreme situation, then there is a use for it.

@frullaccia @producer @DrDave @Manicd @matt_gh2 @brudney @jmatthews @jleo @JayUKB @joethepro @Marathonianbull @rezyserzycia @TrackZillas @alphadawggh2 @driftwood @Gallo22 @Vitaliy_Kiselev @zsound @spreeni @NikolayM @Aashay @caveport @LPowell @maxr @Azo @duartix @yskunto @marciocons @kris @karl @Ralph_B @Alex_K @Daniel_R @cbrandin @all

Hello guys, you all followed the developing since long time ago. Please give some feedback about the seta in last upload in post above (seta inside the_patch_4versions zip), it is the main version with all improvements I could find since the beginning of development. Many thanks.

Feel free to change the things you need to satisfy your taste about the image. You can try copy and paste the scaling tables / deblocking tables from one patch to another. Sometimes this copy/paste works, sometimes needs tweaking in other settings to make it work ok... a patch is a combination of many things working together... I think I already found the combination I was looking for, so I do not want to start another pursuing about patches...

I did a look on it, it uses very low numbers in some position in an attempt to lower the quantization, but to keep a reasonable quantization amount in some position it uses very high numbers wich increases quantization a lot for some pixels, and does not have a frequency similar to the dct frequency.

This is experimental:

an attempt to reduce the green cast in 720p and hbr, and try to make the 720p and hbr textures/noise looks like 24p.

works only for PAL lands, so it is sete = Europe (because the NTSC framerates/datarates does not allow this try)

auto electronic features does not work in 720p and hbr, better not to use native lenses.