Fri Sep 28, 2018 1:14 pm
Frank, that's a pretty bad example.
Try some contrasting detail, with colours as another example, like static frabrics. The monotone pastel red/orange roof tiles shot that way are just hiding issues. Even with your diffusion, just glancing at it there is obvious difference. What I'm going say proves the point. You saved that jpeg at reasonably low quality (like 72-150mb/s h264 in equivalent bitrate, but I don't know what quality efficiency per version of each codec) and frankly, there is not much to compress. But look at the vertical tiles beneath the window, you go down the list and they look less and less real (except the two bottom ones of course). I just looked at the tiles to the left, and saw the adjustment in them first. However, what resolution is this really. The minute miriad of pixel details has the same effect, their preservation on consumer video helps the visual appeal. It is all about how people "feel" about the image, how it subconsciously and consciously strikes the viewers mind. I remember the police interviews in The Bill when they switched over to digital betamax. They used a cinemagraphic lighting style which really brought out the facial details and facial hair, on our lousy free to air broadcast channels. The art of low bit stream encoding, is what to keep, as I pointed out to the industry 14 or more years ago. I recommend ambarella for a reason, before that low bit rate h264 was rubbish from what I saw.
I've sized up the samples at distance to precisely emulate a reasonable feild of view (measured it out to just at the lower end of the optimal feild of view for cinema seating). I can see the differences, but if you don't look for them, the levels are more pleasing at the top sample. You can see the mark in the 3rd or more tile beneath the window in, stands out, while is just gets less noticeable below it. This contrast of detail stands out. You could empty out much of the texture on that mark and it would still look like something realise there (the low contrast between the pixels in there help make it less noticeable) but if you reduced the contrast of the market itself compared to its surroundings, it would look less strikingly real. Thus the levels of the whole pattern of tiles benefit from this in the uncompressed sample. Years ago, when David Newman invented cineform raw, he showed us samples of cineform compared to uncompressed sample. The scene was very much the same, but the contrasts went fuzzy, shallower, much less pleasing in parts. Like cineform at high compression ratios, BRaw seems to suffer somewhat, except what I'm seeing at best quality is sort of what I would expect at 12:1. So, I would hope BRaw will have a CDGN beating mode one day. Something is happening there, we don't know, but maybe it is some sort of sacrifice in debayering to remove noise that some are pointing at.
Now, another issue is the debayer mechanism. It's going to effect the accuracy and level of detail available fur a transcode.
aIf you are not truthfully progressive, maybe you shouldn't say anything
bTruthful side topics in-line with or related to, the discussion accepted
cOften people deceive themselves so much they do not understand, even when the truth is explained to them