[Biro]

From the Photography Life web site...

https://photographylife.com/14-bit-vs-12-bit-raw

[Na Horuk]

Nice read, thanks for posting!
I remember a similar thread from some years ago, about K-5 vs. some other 16MP Pentax with only 12 bit depth. Back then the users said the difference was small, but that it appeared when you overexposed significantly and had to bring the photo down. But I guess the difference is smaller than I thought.
Another thing to consider is that if you shoot raw, you will probably be using powerful software. These days, software can do some amazing NR, contrast editing, sharpening; so i think the difference in bit depth becomes even less important.
That said, a lot of photographers go into overkill anyway. They shoot raw with highest options, use pro software with a big gamut, use all kinds of specialized addons for specific tasks, use super sharp lenses on cameras with really high MP count.. and then upload a 1600x1200 photo to instagram lol

[BigMackCam]

I've read this page before. Interesting stuff, and I have to believe it in the absence of any way to prove or disprove it. What I will say is this... I use Lightroom 6 and Photoshop Elements 14. PSE 14 will only accept 8-bit TIFF files, not 16-bit. Now, when I "edit in" from LR to PSE (forcing a 16-bit to 8-bit conversion), there are occasions where I see differences that look like minor pixelation - particularly in gradual tonal changes of sky and shadow areas. But this is viewing at 1:1 reproduction. Rarely have I seen anything at 1:2 or smaller. Given that, I can fully believe that the majority - possibly the entirety - of photographers (me included) wouldn't be able to tell the difference between 12 and 14-bit RAW

[Tas]

He states that the 12 bit images appear lighter, and this is most noticeable in the low light 'hand' image when you look outside the window. The uncropped image best demonstrates it, particularly when you click on the image to view in the slideshow. The 14 bit image appears to have retained more detail in the highlights, or is this just me seeing this?

[Architecture]

There's some minor hotspotting in the green bush outside the window, but look at the loss of shadow detail of the 14-bit v 12-bit crop of the hand at 3200 ISO. The 12-bit image retains detail in the pillow's pattern that disappears in the 14-bit image.

Would it be safe to theorize that Lr's algorithms while processing the image to a JPG from RAW would affect the DR of the images?

[clackers]

Food for thought, Biro, thanks for posting!

[Simen1]

Back then the users said the difference was small, but that it appeared when you overexposed significantly and had to bring the photo down.

It should be the other way around. Underexposing and bringing it up several steps. Or when pulling up the shadows (actually compressing the dynamic range).

Most system cameras today have 13-15 stops of dynamic range (at the lowest iso, from clipping to noise floor, according to dxo)

Most consumer level LCD screens have only supports 8 stops (bits per colour channel) of (static, non compressed) dynamic range, but some professional screens have 10 or 12 stops that requires special graphics cards (Nvidia quadro or AMD firepro series), appropriate drivers and software support. I believe this is the main reason most people can't see any direct difference without pulling up the shadows or the whole underexposed image.

I don't know the dynamic range of different print methods but I suspect all (or most) methods are more limited then what the camera sensors are capable of.

[Na Horuk]

It should be the other way around. Underexposing and bringing it up several steps. Or when pulling up the shadows (actually compressing the dynamic range).

I think the main complaint was posterization, and a higher bit depth helps with that to some extent. When bringing down blown out skies, every bit of info can help, as that can be the difference between pure white, and just a tinge of blue, a shade of aqua.

[JensE]

I have not shot any examples with both a 12bit and a 14bit camera, but some example pictures where I hit the quantization limit of even the 14bit DNG from the K-5 come to mind. A recently posted one in Project52 was this picture after significant post-processing:

Processing with a flat profile (attached) clearly demonstrates that there is extremely limited dynamic range (significantly less than 2EV) to work with here - basically a thick layer of spray. In addition, I had to underexpose because I could not change lenses (spray and sand blown), the DA 18-135 was already at its widest, I did not want too much blur in the waves and anticipating having to steepen things up wanted to use a fairly low ISO to avoid amplified noise later. This way, I gave up nearly ~2EV at the top end (2bit) by not exposing to the extreme right. That leaves 12bits to start with. Then, considering the main information in the image is spread from around 80% to 130% of the average only, there's another bit gone. Now apply a nonlinear curve to the remaining 11bit of range and you start to see quantization artefacts in the resulting 8bit JPEG image. E..g., I could not get any better contrast in the clouds without posterization.

This shot, taken with 12bit raw data could only be presented with an almost flat sky - very different from what the intention and visual perception at the moment in time was. There was so little color differentiation even with 14bit raw data that I refrained to present it in color altogether.

[hjoseph7]

Great article, sort of reminds me why I'm happy with a 16MG pixel camera as opposed to a 24MG pixel camera, well sort of...

[wombat2go]

I don't think it is very useful to try to illustrate the difference using such low res jpg as examples!
But I agree with the authors comments /conclusions that it is difficult for the eye to discern the increase in brightness resolution,
in a real, properly exposed photo.

That is my experience with the 30 bit Eizo Truecolor monitor here (SX2762W).
I can see the difference on a specially made test frame, but I have no startling revelations that Truecolor
is better for my rather mediocre photos.

The difficulty in doing a comparison here with the Nvidia Quadro switching between 24 bit and 30 bit is that the operating system
must be re-booted each time to run the Nvidia driver to re-set the bit depths in the hardware.
( Incidentally, the Quadro's DisplayPort Truecolor interface can not display the bios or cli during boot
because it has no output until configured by the Nvidia driver)

Camera notes:
The weakest channel on the Bayer is red. For an average exposure on a gray chart, I measured here that the red channel
is about 1.5 stops less than green. ( I put a post on PF some time ago)
So neglecting sensor noise and not being too fancy about adc quantization, on a 12 bit camera, with a stop of overhead in the exposure,
red channel has a little more than 8 bits ( stops) of dynamic range:
( 12, minus one stop of overhead, minus 1 stop of quantization dither, minus 2 stops for the low red level)

So a 12 bit Bayer camera should be just capable of de-matricing to an 8 bit color on a bright section of the image.
But a section in the shade, say 4 ev down, will have about 4 bits of usable brightness information in red channel.
However that would have been known at time of the Bayer design, and it works quite well.