@LPowell - I can see some differences - Flow Motion is a little sharper, and the 444 stuff has somewhat smoother gradations. I think it depends on the subject. The one you show is pretty contrasty and monochromatic. Other subjects may show more or less of a difference.

Your point about matrices not being interchangeable is well taken. Changing matrices changes everything else too, and it's a lot of work to get right!

hello... i needed to see for myself, so here is my comparison of the two matrices:

flomotion 2.02 & Valkyrie '444 Soft',

both 3gop 24p high bitrate patches

if you were to zoom into details (such as the ingmar bergman box set, or butch cassidy and the sundance kid case...)

u can see how the two are different, hope this helps!

24p mode, 14mm@2.5, 400iso, cloud WB,

Lpowell, regard the scaling matrices, I wasn't aware of that. Thank you for the information. After trying both settings my self I see advantages and disadvantage on both FL and soft 444 so to speak. But I some how tend to like the feel of AN 444 soft. btw, the fighting comment was not meant to you alone, so no hard feelings. 

Flaschus, Some how the cables in the background seem overly sharpened on AN? Tend to like the FL in that regard. Than again.. There seem to be less noise in the shadows regard the AN. Confusing stuff... 

An avid lurker here -

Can somebody point me to the discussion where it seems to have been explained that 25p is really 25i ?

@lpowell - I looked up the reference you linked. My formula calculates what he calls QP. It's a little confusing because that number is a product of various parameters included in H.264 headers and the scaling matrix. And then there is PTool's "Quantizer" Parameter. Let's agree that the index into the table is "QP", the indexed entry is "Q", that PTool's parameter is "Base QP", the frame's base QP is "Frame QP", and that the macroblock QP offset is "macroblock QP offset" - that should eliminate the confusion.

EDIT: VK doesn't like deleted posts, so PLEASE NOTE: The conclusions below were based on artifacts introduced by the photo editing software in the course of combining two images. KINDLY DISREGARD.

==============================

The two Flaschus shots are not quite identical, which may have influenced the result, but if you put them side by side, the most glaring difference is in the carpet -- it's a darker shade in FM than Valkyrie, and FM appears to pull out more detail, maybe for that reason. The difference isn't subtle.

When you start raising the blacks to get shadow detail, there appears to be more noise in Valkyrie. Push it to an extreme, and there's more apparent macroblocking in Valkyrie.

Again, am not taking sides, just trying to use my eyes. Maybe the engineers and colorists among us will have more useful or valid observations.

@Zaven13 Thanks for the comparison shots. Something looks amiss in the last shot, however, FM_24H-140zoom_freeway_ETC, that's not present in the other shots. Atmospheric distortion from the auto exhausts, perhaps?

@LPowell - Most of the decoders I've seen take the PPS QP value, add the macroblock QP offset, multiply that by the matrix entry, shift right 4 (divide by 16), and use the result to look up the Q value from the quantization table. This is done for each matrix entry in the frequency domain. This is a somewhat simplified explanation as the PPS QP entry is actually the QP value -26 (I don't know why they did this). Offhand, I don't remember the exact sequence, but it's close to what I described. The JM decoder probably doesn't do it that way because it is not optimized at all - but practical decoders typically work essentially that way. The design of the H.264 standard takes 16 bit processors into account so that CODECS can be as efficient as possible.

So, to recap - A frame QP value (frames in the GH2 are single slices) is added to a QP offset for a macroblock, that value is multiplied by the appropriate quantization matrix value, the product is divided by 16, then that value becomes the index into the quantization table.

Onionbrain (DrewNet) made some big statements both in his original promo video of Apocalypse Now and here in print http://www.personal-view.com/talks/discussion/comment/81924#Comment_81924 "As a side note -- having viewed this stuff on calibrated production monitors and large screen projections -- I will tell you and everyone without qualification that this is the best looking footage I've ever seen from a hacked GH".

I'd like to see actual evidence of what OnionBrain is seeing here to backup these claims. If you are going to be that outspoken on promoting this patch, then you owe it to the lesser experience users here (myself included) to provide such proof. I'm not discounting the work people have done here, but we need to see what you see rather than just take your word for it, else it becomes just hype.

How about comparing it to Sanity as well since there's a big difference, filesize wise?

@Zaven13 Yes, thanks again for your test! Southern California, that's what I thought.

@SuperSet. Good idea. You lucky guy... This forum just nominated you to do that testing...