Is there such a thing as 8 bit raw? I like 10 bit the most so far because of the resolution gain. Just wondering if 8 bit could get us even more resolution and dynamic range than h.264? Or would it basically be the same thing as h.264?
Did you post this to Duplicate Questions or did a moderator move it?
I wonder about that too. 8-bit uncompressed is pretty good and doesn't have the artifacts of H.264.
I think it got moved. Maybe it's been asked before? I've never seen it on here. I started to ask about it in the mv1080 topic but that was off subject so I did this post..
8-bit linear data doesn't make much sense; maybe only at very high ISO. 10-bit linear is already showing its limits (http://www.eoshd.com/comments/topic/21004-12-or-10-bit-raw-magic-lantern/?page=3#comment-168058). See also a 8-bit DNG created in post here (https://www.magiclantern.fm/forum/index.php?topic=18323.0).
On the other hand, 10-bit raw (Bayer) data with a nonlinear curve (https://files.apertus.org/AXIOM-Beta/optimal_curve.html) applied would be nearly lossless on current sensors, and 8-bit nonlinear would be very close.
The only issue: how to configure Canon's image processing modules to do this stuff?
BTW, recovering the scene-linear data from a clean HDMI recording (or even from a JPG or H.264) could be interesting (one just has to undo Canon's color science). It's already done, at least to some extent (lookup e.g. radiometric calibration).
I got a feeling it might be kinda cool.. I saw a build before that had 10,11,12,13 and 14 bit. I'd love to see one that said 8,10,12,14. If it's possible.. I was thinking if the dynamic range is still good and the sharpness is good I might just through it into black and white if the color isn't so great or there's a lot of noise.
11 and 13 were never possible. They were in the menu list because of my laziness and a quirk of the menu API. You couldn't actually select or use them. It's very easy to add 8 to the list, making it actually work is another story, its not clear what one would pass to the current method to indicate 8 bit or even if that's possible (currently 10bit is selected with 0). There may be some separate module, idk.
I do suppose getting the hardware to give us 8 bit would be very useful for in camera preview and analysis (histogram and other raw overlays).
Oh ok. That's a pretty good explanation. What's less than zero? Lol