It appears that there aren't even any samples (that we can see-not graph plots but images) that demonstrate the difference of "DownBitting" from 16 to 8 bits.
The difference is most likely to appear in low light areas such as deep shadows, where there are fewer data bits available to cover the tonal range than at the bright end of the scale. You won't notice anything in good lighting.
Chances are that the middle tone of the raw image data is now much higher than it was on previous cameras (to suit the higher maximum data value that is available) and that will leave up to 4 times the number of tonal values in each ev interval or zone in the darker areas.
It will also be most obvious if there is not too much noise burying the data in those dark areas, so perhaps only or mostly in the noise reduction mode. People don't like using this mode because it severely limits the available shooting buffer space.
I have downloaded a 1D3 raw file and compared it with my 1D2 raw file in the DPP raw image histogram.
The 1D2 has an exposure range scale of about 12 stops (12 stops is the theoretical limit for 12 bits) covering -9ev to +3.6ev (approx).
The 1D3 has an exposure range scale of about 14 stops (14 stops is the theoretical limit for 14 bits) covering -11ev to +3.7ev (approx).
This indicates to me that most if not all of the extra data is held in the dark end of the range rather than the bright end, although that might change for a high tone mode shot.
For the 1D3 the maximum tonal level is near enough 2^14 or 16,384. 3.7ev less than that is 1,260. So canon have defined middle tone to be 1,260 on the 1D3 (if my graph reading skills are correct. It was easier to determine on older versions of DPP but they cannot read 1D3 files). One stop darker is half of that or 630. Another stop darker is 315. Then 158. Then 79. etc. SO between three and four stops darker than middle tone there are (158-79)=79 tonal levels available to describe the scene.
Note: the middle tone is the one that maps to a value of 127 or 128 on an 8-bit data file when there are no additional exposure adjustments being made. The maximum value will map to 255, as may many other values depending on contrast settings. 0 will map to 0. Everything else depends on the shape of the tone data conversion curve used by the software and/or camera.
On the 1D2 the maximum data value is about 4096 and middle tone is about 338, so one stop darker is 169, 2 stops about 84, 3 stops 42 and 4 stops 21. Thus there are only 21 levels to describe the scene between 3 and 4 stops darker than middle tone, or near enough a quarter of what the 1D3 has to offer.
So it is definitely in these dark areas that the extra data capcity of the 1D3 will be evident provided the noise doesn't dominate.
I almost forgot to add that in an 8-bit file these darker areas have almost no data values allocated to them so the difference between a 1D3 jpeg and a 1D2 jpeg will be practically non existent in terms of any benefit derived from having 14-bit raw data unless you need to do some serious exposure or contrast or curves adjustments prior to conversion from raw.
There seems to be an expectation about finer tonal levels than 12bit. It will Not be the case automatically. What matters is where the sensor noise level is. With 1DsII, it has 9 stops of DR. In other words noise, level is 3 bits. If you increase the data size to 14bit without improving the noise, you get the same 9 stops of DR with 5 bits of noise, which is basically the theoretically minimum tonal resolution possible.
So, the effectiveness 14bit will be realized only if the noise level has been improved. How is it compared to 1D2 in terms of the noise ?
Alan321 wrote:
The 1D3 is said to have at least a one stop advantage over the 1D2 without using special noise reduction mode.
1 Stop is not enough because we were utilizing only 9 bits of the 12 bits
2 Stops would start challenging the 12th bit.
I speculate that the 14bit may be useful for in-camera RAW data manipulation such as HTP. Sentimentally, I don't like this notion, "Your RAW data is not raw anymore". Leica M8 did it to the extreme already
Pondria wrote:
There seems to be an expectation about finer tonal levels than 12bit. It will Not be the case automatically. What matters is where the sensor noise level is. With 1DsII, it has 9 stops of DR. In other words noise, level is 3 bits. If you increase the data size to 14bit without improving the noise, you get the same 9 stops of DR with 5 bits of noise, which is basically the theoretically minimum tonal resolution possible.
So, the effectiveness 14bit will be realized only if the noise level has been improved. How is it compared to 1D2 in terms of the noise ?
I haven't seen other owners disputing that noise levels have been improved, what makes you think that is the case? The question seems to be how much have they been improved. But even if the noise level improvements are minor, they do allow the 14 bit files to exhibit finer tonal gradations in looking at RAW files from 1DIII, 1DII, IIN and 1DsII. As has been pointed out, you aren't going to see anything in compressed jpegs posted here unless a lot of work is done in PS to crops to specifically call out the areas for web viewing. That sounds like something you need to do yourself.
I almost forgot to add that in an 8-bit file these darker areas have almost no data values allocated to them so the difference between a 1D3 jpeg and a 1D2 jpeg will be practically non existent in terms of any benefit derived from having 14-bit raw data unless you need to do some serious exposure or contrast or curves adjustments prior to conversion from raw.
So basically the 14-bit files are worthless unless we work in a 16-bit pp environment? Am I incorrect?
No. When converting to an 8-bit file from a higher bit depth you can select what range of the data you want to map down in to the 8-bit space.
Also, you can't really see 16-bit color on a computer display because the displays are themselves only 8-bit. You'd have to print to see the difference.
If you had a 16-bit file and mapped the full data range to 8 bits, the difference in your print output would be that the 16-bit image would have finer color gradation.
Pondria wrote:
There seems to be an expectation about finer tonal levels than 12bit. It will Not be the case automatically. What matters is where the sensor noise level is. With 1DsII, it has 9 stops of DR. In other words noise, level is 3 bits. If you increase the data size to 14bit without improving the noise, you get the same 9 stops of DR with 5 bits of noise, which is basically the theoretically minimum tonal resolution possible.
So, the effectiveness 14bit will be realized only if the noise level has been improved. How is it compared to 1D2 in terms of the noise ?
...Show more →
It has finer tonal levels. Bit depth limits the maximum possible expression. But we don't know that cameras before the 1DIII were able to meaningfully express more than twelve bits. We don't know if the 1DIII even needs 14 bits. All I know is that there are finer tonal levels in the 1DIII than previous cameras. Unless you have the means of measuring each part of the system there's nothing to say except the camera has finer tonal levels. Resolution interacts with bit depth, but I don't see that discussed. 8 bit hi res high end drum scans look the same as 16 bit hi res high end drum scans. How is that?
Alan321 wrote:
...I almost forgot to add that in an 8-bit file these darker areas have almost no data values allocated to them so the difference between a 1D3 jpeg and a 1D2 jpeg will be practically non existent in terms of any benefit derived from having 14-bit raw data unless you need to do some serious exposure or contrast or curves adjustments prior to conversion from raw.
- Alan
Great answer on the whole, Alan. Your final passge put the matter in perspective: where the extra 2 bits of depth would be useful - even in theory - is in making radical contarst adjustment within the shadow range.
Do we know whether the medium format digital backs are even making use of a full 14 bits? I have seen that called into question in some places. My gut feel is that some are and some are not.
The information that I have seen from a Kodak engineer, who has said that you need more than xx decibels in terms of the signal to noise ratio (I will try to find the specific number of dB) to make use of 13 bits, then xx for 14, etc. Pretty similar tro the analysis presented here.
As mentioned, there are a lot of other factors that can affect image appearance. Canon made huge improvements to the look of their files from the 10D to the 20D when they moved from th Digix 1 to Digix 2 processor. No increase in bits in the A/D converter, but a huge inprovement in other areas, including noise (which may also be the sensor, etc.)
There's more tonality below middle gray. It's more than just the darkest shadows. With wide dynamic range shots don't block up as much. But I hate for reality to get in the way of math-based internet conjecture. Carry on.
dcmiller wrote:
There's more tonality below middle gray. It's more than just the darkest shadows. With wide dynamic range shots don't block up as much. But I hate for reality to get in the way of math-based internet conjecture. Carry on.
From my understanding, and from what I saw Canon explain, is that the 14 bit will effect shadows and, more specifically, color rendition. I remember seeing a sunset shot in 12 bit and again in 14 bit, and they showed a much smoother color gradient that the 14 bit had over the 12 bit. Again, I'm not an expert on this aspect of things. And going to Medium format compares you to 48 bit, correct? Not exactly a fair comparison. It's funny, mmurph, but I found my 10d files to be far suprior to my 20d files. Mostly in the shadows when shooting ISO100. Yes, the ISO800 is far better in the 20, but when doing some serious post production work, I found the 10d to be the favorable image.
