Dmitry Shijan wrote:LOL. Sensors that can capture and process image in 16 bit are very rare. Those mostly single pixel line array CCDs from some high end film photo scanners. They are extremely slow and produce a lot of heat and consumes huge amount of power. 12-bit represent 4096 different numbers,
16-bit represent 65536 different numbers. Huge amount of data to process. Not sure if CMOS 16 bit sensors ever exists.
12 bit in single gain mode is more than enough as well as 2x11 bit combined in dual gain mode. Sensor bit depth capture and internal processing is hardware limit, it is not the thing that BM developers can adjust. In real life going from 12 to 14 bit sensor internal processing will not brings a lot of additional quality. Sensor dynamic range is way more important than 16 bit internal processing.
There is a dude on Youtube who makes a lot of side by side tests and shares links to original files in comments. So he compare P6K vs some modern RED camera. When i open both BRAW and R3D files in original 1:1 size in Resolve i was shocked that BRAW footage has almost no aliasing artifacts and same time RED footage produced a lot of aliasing artifacts. So BRAW pre-processing is not a bad idea. It just needs some improvement for compression. And also it need to allow user to adjust noise reduction strength. Also it need to allow to switch pre-processing off for situations when you use OLPF filter and no need software anti-aliasing processing. Same option may be usable for ProRes.
Thanks Dimitry for the Braw recommendations. That is what I have been saying.
There are various 14 bit mode Sony sensors in the consumer range. Just BM leaves things clouded. Having 12 bit log to move the exposure as a reason, seems not as good as having 14 or 16 bits sensor values packed into 12 bits log. But, they have left it hanging. Why does 16 bit generate more heat then 12 bits Dmitry? I would expect more, but not that much.
I think Red is a bad example to compare to. Your artifacts will depend on your olpf your choose, and they go to some ridiculous amount like 15:1 constant without the level of noise reduction Braw does (as I told the industry years ago, noise reduction is the key to higher intra compression ratio, and you would get camera engineers debate the use of it). So, above 6:1 I wouldn't trust it, maybe less on the original red code. 3:1-4:1, Braw 2:1-3:1. Which compression ratio, codec versions and test samples was he using there? But, I specified various noise reduction techniques (though I don't remember how many new ones I kept in notes. If Red or Sony started doing it on sensor after my suggestions back then, I probably suggested it). Anyway, the point was, that I would have put spatial noise reduction do far down the list, it would have been off of it. I would have put good temporal at the top, though I had some interesting techniques in between). Elsewhere on the forums here, I've laid out my second generation image preservation preference list. First one did major improvements, but this one was more nuanced for high quality imaging. But, in that you can compress the parts which are sky or plain more, hair less, face in between, etc, dynamically for more compression while maintaining up to lossless quality. This is IS (intelligent Structure) programming rather than AI, to optimally process the image compression by rule set, or dynamic rule set, and dynamic analysis if needed.