Quote from: names_are_hard on May 13, 2023, 12:23:00 AM
I can't answer all of these but I can give you a start.
I don't think anyone knows if 10-bit h264 is even possible, or anything above 1080. I suspect not, very sure it's not possible currently.
Binning is done at hardware level by Canon. It collects light from multiple sensels but sends back one measurement, allowing collecting light from a larger portion of the sensor for a lower bandwidth, at the cost of spatial resolution. I don't know how the sensel data is combined, some kind of average I assume but I don't know which one would make the most sense. Geometric mean possibly?
If it's clipped, it's clipped. Different programs will use different algorithms to try and make the clipping look nicer, but the information is just as lost. You can losslessly convert mlv to dng and compare recovery in two tools. I haven't used MLV App, this question is probably better asked in an MLV App thread, maybe here: https://www.magiclantern.fm/forum/index.php?topic=20025.msg243653;topicseen#msg243653
You might get better feedback about the video if you ask here (and explain you want review / feedback): https://www.magiclantern.fm/forum/index.php?board=19.0
Thanks! I will be checking out those forums to see what more I can find. I appreciate you taking the time.
In the meantime, I did a little messing around with various settings outside in the parking lot of my studio. I shot the same shots in various formats including the 6D's own h264 format for comparison. Everything is shot on my Contax Zeiss 28mm at about f/6.3 with a VND filter at 1/50 shutter. All 24fps. All 100 ISO. Nothing was changed (not even focus) and everything was shot in the span of 10 minutes. RAW shots were processed to ProRes 4444 in MLV App with ARRI LogC and then the LogC to 709 LUT was added in Premiere.
4K 1:3 Crop video: https://www.dropbox.com/s/is50a52h4r13nk7/6D%2028mm%204k%201x3.mp4?dl=0
1080 various formats compared video: https://www.dropbox.com/s/idiht9wwwk6ur44/6D%2028mm%201080%20various.mp4?dl=0
Initial findings (I would love anyone's feedback to see if I'm missing something):
1. Raw mode without crop mode enabled is far superior to the Canon h264 (not surprising). Rolling shutter is about equal in both modes and very usable and maybe a little less noticeable in h264. Moire and aliasing is horrendous in both but worse in h264. The power lines and card edges render poorly. The patterns from the shop roll up doors is all over the place. Again, all worse in h264. I wish I would've used a flatter profile in h264
2. 1:1 crop mode is the cleanest but the crop factor makes even my 28mm telephoto. that said, there's practically no moire or aliasing. very clean lines even on power lines. At 100 ISO, the noise in the image is negligible but I would be curious to see how the noise patterns did in the dark at higher ISOs as compared to a mode where more of the sensor is utilized. Rolling shutter is insanely prevalent.
3. 1:3 crop is surprisingly good image wise. If you look at my reference videos, the power lines render really well even when magnified to 100% in the 1080 video I exported. You can see some softness in how the power pole is rendered as compared to the 1:1 cropbut this is surprising good!! The results are astonishing. That said, rolling shutter is increased intensely. Is this due to slower sensor readout or because we're actually recording less horizontal pixels?
I have the Sandisk Extreme Pro 170MB/s card and my SD overclocked at 192 but I can only get about 23 seconds of recording time on 1x3 crop mode. Does anyone have a solution to this?