Paul,
As I said above. Problem is not ADC but the data volume through the pipeline. 1DsII is limited to 4 FPS because of the bandwidth of the data transfer. 6 bit data ( instead of 12 ) would have made it a 8 FPS.
Paul you are right about that, but having a lower number of bits for each pixel will reduce the amount of data (megabytes not megapixels) that needs to be exported from the sensor to memory, and will allow for a faster frame rate to be used.
I haven't sat down to calulate the difference in file size, because to me it really doesen't matter. Put a larger battery in the 1Ds and 5D series and run the camera processor faster. From what I can find out they are only running about 20Mhz now. Not being a PJ I prefer the detail at the expense of throughput. 1 frame every 2 seconds is fine by me. Computers are cheap these days. With processors like the AMD 6000+, and the Intel E6600 and Q6800 processing speeds are getting fast enough to keep up with 30 Mb files and 300Gb hard drives are under $100.00 now. The new processors coming on line in the Oct-Nov time frame are going to more than handle these size files, especialy if Adobe gets off the dime and releases a 64 bit Photoshop to match the current hardware. Adobe caught a lot of flack for not releasing a 64 bit version this time. I'm still P.O. over this as it solves the 3 Gb memory limit of Windows.
The 32 bit Adobe actually runs slower on a 64 bit machine because it has to run in the emulator.
Jeff and others,
I'm working on the 1D3 raw file that Jeff sent me. I found that my old DNG reader had problems with the new DNG file from 1D3 CR2 file. Sorry for the delay. I'm working on it.
You might want to check FredMan's offering of two RAW files (m3 and m2); he shot these at my request to determine if Mach banding would be minimized as a result of 14-bit processing. Here is a link to the page that will get you to his post at the 1D-Series Forums.
I downloaded the files and could see no difference in banding between the two cameras (I really can't believe there is no visual difference in the 16-bit TIFF files but this is what I found). Until I saw his RAW files I was absolutely convinced Mach banding would be reduced by a factor of four; I'm a bit bummed out but this find but it isn't the first surprise I've ever encountered in my life. :-)
I've finally been able ti dig into some of the RAW files, and it occurred to me that I might be able to illustrate a 'typical' scenario where 14-bit source files could be a benefit: an extreme gamma adjustment to darken skies. This often occurs when doing B&W processing for landscapes, trying to mimic color filters. I've always had trouble with skies posterizing in these circumstances, so thought I'd try it on a bird image.
The left side is the in-camera 8-bit JPEG straight from the camera, the right is from the CR2, converted in DPP. Keep in mind this is by no means scientific, and honestly some of you more well-versed might have to interject whether this is likely more due to the CR2 conversion to 16-bit than anything.
Anyway, here it is (notice the difference in posterization):
You might want to check FredMan's offering of two RAW files (m3 and m2); he shot these at my request to determine if Mach banding would be minimized as a result of 14-bit processing. Here is a link to the page that will get you to his post at the 1D-Series Forums.
I downloaded the files and could see no difference in banding between the two cameras (I really can't believe there is no visual difference in the 16-bit TIFF files but this is what I found). Until I saw his RAW files I was absolutely convinced Mach banding would be reduced by a factor of four; I'm a bit bummed out but this find but it isn't the first surprise I've ever encountered in my life. :-) ...Show more →
Just saw your recent post when I posted my last to this thread. Note that apparently you cannot log in to your linked website unless registered (which I chose not to do). Thusly, I have no idea the content of your link.
Jeff wrote:
..., and honestly some of you more well-versed might have to interject whether this is likely more due to the CR2 conversion to 16-bit than anything.
I also think along that line. Actually, you can also try the 8 bit conversion and see if the banding comes back.
Jeff wrote:
Just saw your recent post when I posted my last to this thread. Note that apparently you cannot log in to your linked website unless registered (which I chose not to do). Thusly, I have no idea the content of your link.
Jeff, sorry! The gentleman that set up the board, Wally Jarrett, is a pro sports shooter using Canon gear and apparently that is how he wants to run things. The boards were created to strictly cater to owners or folks interested only in Canon's 1-Series cameras and associated accessories.
I got an opportunity to look into the 14bit RAW data that Jeff sent me. Thank Jeff ! Those are the files that he posted on the forum, the high DR scenes. One of them was HTP enabled. I am going to share what I’ve found for last two days. This seems to be just one baby step toward more study.
I used several different sw tools including my own. The work-flow is as follow.
1. Convert CR2 to DNG
DNG is open format with documentation and tools. It’s a lot easier to get to the RAW data as the sensor captured. I used Adobe DNG Converter 4.1.
2. Extract raw RAW from the DNG
By RAW, I do NOT mean the RAW that ACR or DPP presents to us. This is sensor data, as is, represented as 16bit gray TIFF. Adobe provides free DNG SDK with source code. It comes with nice utility called DNG_Validate.exe. It produces the data in various stages of the process. The 1st stage is no-mosaic'ed, unmodified data. If you load it on Photoshop, you can actually see the Bayer patterns like this http://www.sesee.com/Photo/Exports/1D3bayer.jpg
One interesting thing is that the actual sensor is larger than the image size that it produces. The border padding is larger than you may think. For 1D3, it is 3984x2622 ( vs. 3888x2592 image size )
3. Data histogram
You need a program that can count or produce histograms from the data. I used an old Matrox MIL utility called Inspector. And I feed the Inspector data to my own utility to produce the histogram stat in Log scale. Then it is finally copied over to Excel for better presentation.
Let me explain how to read the table. The first columns ( Stop ) covers the entire 12bit or 14bit data range. The numbers are in stops. For example 8 covers the range of values starting from 2^8 to next ˝ stop or 256 to 361. The 1Ds2 column is for 12bit data from 1Ds2. And the next two columns are for the 1D3 data from Jeff. You can easily verify that every single pixel is accounted for by checking that the sum of the numbers in each column equals to the # of pixels the camera produces. http://www.sesee.com/Photo/Exports/1D3-raw.gif
4. Conclusion
(1) Yes, 1D3 produces 14bit numbers ( maxed at 16383 ) while 1Ds2 maxed at the largest 12bit number (4095). BUT this is as far as the 14bitness goes. You probably noticed that there is no data captured in lower bit range. The real data captured are just about 5.5 stops for 1Ds2 and 4.5 stops for 1D3. There seems to be no other difference between them except that one data is shifted by two bits. The darkest dark that you can go down seems to be limited at ~ 100 for 1Ds2 and ~1000 for 1D3. ( This may be something to do with ISO. 1D3 was from ISO 200 while 1Ds2 was from ISO 100.
(2) What Rico said may be true – “Due to internal reflections of optics, max DR that we can have in the film chamber is limited regardless of the sensor”.
(3) HTP doesn’t seem to have any effect on the RAW data range. I strongly believe that HTP is done in post processing, being in-camera or RAW converter.
(4) In linear scale, far larger data range is allocated for higher stops. But see, how densely the data are concentrated in just a couple of stop span.
Edited by Pondria on Jun 24, 2007 at 11:47 AM GMT (Reason: David's data added)
I wonder what the chart would show if you shot one image of a white wall just below clipping and one of a dark room with no specular highlights at iso 6400?
