Steve Fishwick wrote:Whilst there is banding as Robert has shown, I am not seeing it anywhere near as severe, even with a quick and dirty sat and color boost pushed right up. This is in my work online suite with a 10bit grade A reference monitor. The whole chain has to be considered when you are faced with artefacts: this is coming from 8K/12K down to HD, and then here the stills are further compressed to get below 1mb. What is the bit depth, sampling rate and codec of your timeline/output?
Steve - I can clearly see the banding on my calibrated 10bit UHD EIZO monitor, in an UHD timeline.
It can be seen in the 16bit TIFF files I rendered out.
As I have shown, the banding is inherent in the Red channel of the BRAW footage.
Steve Fishwick wrote:Since however, neither the 4.6K nor 12K camera, with 12bit Braw should exhibit banding, this is why I suggested aliasing. Aliasing can occur through the concatenation of codecs in post and it can often look like bit starved banding. I see it quite often in my job as an online/grading editor in UK broadcast; especially from very high rez acquisition sources, to HD, that we still broadcast in here. So therefore, one explanation here could be that the 4.6K Ursa is producing significantly less ultimate aliasing over codec transition than the much higher rez 12K. And could be why the 12K needed an OLPF before Netflix acceptance; and the 4.6G2 didn't, for example
I don't think any of these valid points is here the case.
It looks like a problem created by the partial de-mosaicing of the RGBW sensor in camera, an edge case for the algorithm. I think we should point Kristian Lam and Hook onto this for further investigation.
Uli Plank wrote:If the banding is getting worse down the line, adding very mild grain should help. Sometimes, footage out of the cameras is just too clean, and any kind of compression (even BRAW) is reducing the information in very subtle transitions.
If you look at the samples you can see they are far from being too clean - in fact the colored noise is quite strong.