Boson

apefos

I have a better understanding about the gop tables now, far from dominated, but good notions.

advice for PAL shooters: boson is not tweaked yet for work in PAL, I am doing attempts on it.

apefos

I have good news, I am using the weekend to try to map and understand the GOP TABLES.

all the secret for image encoding is here, the magic happens here.

of course the magic in gop tables happens after finding the best for all other settings, which is already done.

apefos

these are the sete and setf with higher datarate for 24H 128000000, the 24L is now 96000000

the gop table for 24p is the panasonic stock: "1080p24 GOP Table=1, 3, 0, 0, 8, 0" this gop table makes the I frames very high when there is lots of details in the image, but when there is no details the I frames decrease in size due to the variable bit rate.

to implement the idea of keep the I frames always very high (around 1.200.000) and the B frames always very low (around 300.000) I think it will be done tweaking the gop table above.

I do not know if it is possible to turn the codec into constant bitrate, but at least to keep the I frames as big as possible to record as much detail as it can, this will make the steady low iso shoots more sharp and detailed due to less compression in the I frames.

apefos

@producer

if you or someone else would like to tweak the e and f versions this is the way to do:

the versions e and f are an attempt to keep the I frames very high and the B frames very low in 1080p24, this makes the image more sharp and detailed for steady shoots. this is done in the gop tables. the gop tables are the stock in these patches, I did not find a way to avoid the automatic ratio variation yet.

In my math calculations, eight I frames at around 1250000 size each and sixteen B frames at around 300000 size each will deliver a great sharp and detailed image. the compression in the I frames will be only 2.5 and in B frames will be 10 which is in a excelent low value for a B frame when the I frame is big.

the datarate will be around 128000000, so to tweak these versions there is the need to change the 24H datarate to 128000000 and the 1080p24 top limit to 128000.

the maximum limit value for an I frame is in the sweet spot with a 2.5 compression ratio which allows this size, so now it is a matter of learn how to manipulate the gop tables, but I do not know how to do this yet, needs tests.

the correct numbers in gop tables will allow the I frames to keep in 1250000 size all the time and the B frames to keep at 300000 size all the time. this will deliver the sharpest and most detailed image for steady shoots.

after you find the best settings for each hack patch, all the secrets are in the GOP TABLES, they manipulate the size of the I, B, P frames and how much they will vary in size depending on image details, and consequently the ration between them. knowing how to do this makes possible to extract the best form the datarate to get what you want from the image. mainly you can have priority to sharpness and detail or to fine random noise, or an automaticaly adjust between both. for cinema a constant fine random noise is better, for advertising a sharp detailed image is better.

I think that the scalling matrices and deblocking matrices are important, but after you have a good matrix then the gop tables is what does the job.

these last patches I uploaded are very careful tweaked in all settings, they are a perfect base to play with gop tables and understand what they do. a good thing to do would be to map the gop tables, starting with the panasonic stock values and increasing or decreasing each number one step each time. if we map the gop tables, we will can do whatever we want with the image.

at this moment I cannot do this because I am full of other tasks, very busy...

apefos

@producer

I think the versions e and f do not make sense for cinema production. e and f use the panasonic stock gop table for 24p and this makes the I frames to change the size: when there is low detail with noise the I frames are small and B frames increase in size, in the death charts which are full of detail the I frames are very big and the B frames very small, so I perceived that I can see more details in the death chart. But I can see this with the eyes very close to the screen as a pixel peeper seeking for perfection. before answer your question I was thinking about if e and f would be useful someway, but I perceived that I does not deserves changing firmware. If the gop table was tweaked to keep the I frames very big all the time so yes it would be more useful for a very sharp image, but I did not tweaked the gop table this way and the behavior of I frames size is automatic in the e and f, I will not use them.

for cinema production it is better to have the same random noise no matter the light or details of the scene, so the other versions deserves the use, they are careful tweaked for the purposes described in the first post.

at this moment I do not have enough time to tweak the e and f to keep the I frames very high all the time. this would make them useful for good light use without noise, with the camera quiet, to get a very sharp and detailed image for pixel peepers.

producer

Would you show few tests of sete and setf in 24H/24L??

Howdy, Stranger!

Categories

Tags in Topic

Top Posters