5D Mark III High-ISO raw files

[SambalOelek]

Since I saw a couple of people requesting RAW files, I thought I'd share a couple. I would also appreciate any advice on how to do anything useful (e.g convert) with them (They can be displayed just fine in ZoomBrowser EX):


I pretty much picked a file hosting service at random, so hopefully downloading them will be not be too much of a hassle.

5D3 RAW @ ISO 51200
5D3 RAW @ ISO 6400
5D3 RAW @ ISO 12800

I have a few more samples, shot at ISO 102400, 25600 and 12800 as well, if anyone's interested. Apparently, the firmware version of my test camera was "4.9.2"

Edit: Added ISO 12800 sample

 

[blufox]

It asks for login.
Can you please host it on a site from where we can download without registering.

Would love to work with RAW file in DPP.

Cu,

 

[SambalOelek]

It asks for login.
Can you please host it on a site from where we can download without registering.

Roger that, I have uploaded to a different webhost. Feel free to rehost if you know a better way to distribute 40+ MB files.

 

[LetTheRightLensIn]

Do you happen to have any black frames and white frames at ISO100?

If you set camera to 1/8000th, put the lens cap on and snap a shot and then set it to an exposure long enough to blow the image out to pure white, all channels blown, over a large chunk of the image, from those two shots, the RAW files can be analyzed and you can find out the SNR, DR and low ISO banding characteristics.

That could answer the question as to whether they have approached Exmor sensor dynamic range or not.
And whether they have fixed up the pattern banding some.

(Assuming software such as IRIS would be able to at least partially open the files.)

 

[altenae]

Now we need the raw converter..........come on Adobe.....just kidding.

I have looked at the 6400 ISO raw file and (to me) it looks much better than a 6400 ISO raw file from the 5D MK II.

Edward

 

[epsiloneri]

Mmmmhrm... is there a version of DPP available that can handle the 5D3 raw files? Not my version 3.11.4.10, apparently.

 

[altenae]

is there a version of DPP available that can handle the 5D3 raw files?

Yep....Canon has the version....
I am afraid it will not be released before the first 5D Mark III is available for sales

Edward

 

[photogaz]

Is not possible to hack DPP or Camera Raw into thinking its opening a different camera model or are the output formats so different from each model?

 

[nitelife2]

Rawtherapee can open them (at least the linux version). The ISO6400 noise of the 5D3 looks like ISO100 of my 7D. I am really impressed!

EDIT: Appended screenshot.

 

[altenae]

Yep tried it also with Rawtherapee, but the colors are not converted correct.

The browser shows better colors:

 

[nitelife2]

Yep, the colors are wrong. But I can be sure that no noise reduction has been applied. Chroma and luminance noise are very low IMHO.

 

[MaGiL]

looks very good in my opinion...

 

[swampler]

I have a few more samples, shot at ISO 102400, 25600 and 12800 as well, if anyone's interested. Apparently, the firmware version of my test camera was "4.9.2"

I'd like to see them if you get the chance to upload them, especially the 12800 and 25600.

 

[Tijn]

IrfanView opens them for me, too. Some free lightweight image viewer for windows. (See www.irfanview.com )

The 51200 sample looks really weird, though. The noise seems to be present in relatively massive same-coloured blobs that are easily 5x15 pixels in size. Noise reduction doesn't help much against it (I converted to TIF with compression set to "none" in IrfanView, and then imported into LR). See included 400% crop...

I hope this is some error caused by my way of dealing with the raw file. Of course this is just the first ISO range "expansion" and not much should be expected, but this seems pretty bad.

(Edit: probably a RAW conversion error.)

 

[Radiating]

I'm not sure if this is the raw converter (rawtherappe) or the camera, but I see noise reduction being applied even without any noise reduction set. At the sensor level it looks like the camera has about a 1 stop improvement, which is a truely epic acheivement because only .5 stops were expected if they were really trying to focus on lowering noise but then there is noise reduction on top of that which you can't remove and that leaves fewer options for NR.

 

[altenae]

Not much we can say about it without the correct RAW converter.

Also we are talking about 51.200 ISO !!!!!
If you expect usable images ????

Be realistic.

 

[Tijn]

The poor quality images I produced are probably in large part due to errors in the raw conversion. When looking at the raw file in IrfanView zoomed in, it seems to display it in clusters of 8x8 pixels that seem to get some of the colours right, but these clusters have contrasty edges opposite to one another (making them visible).

Phew.

 

[qwerty]

It took me a few minutes to do the raw conversion, and I had problems with a magenta cast to images, so hopefully I can save the rest of you a bit of time.

I am attaching 100% crops as PNG files. I used shotwell (under linux) to convert to a tiff, then LR3.6 to crop and export as a jpeg (I tried to post a .png unsuccessfully earlier). As mentioned previously, it is possible that a better raw conversion would result in better quality; however, I have a 5d iii on pre-order and am trying to be realistic and restrain myself from wishful thinking.

Does anyone have a similar 5d ii shot to compare to? I would be very interested in seeing a side-by-side with a 5d ii shot at ISO 6400 and 3200 (but I only have a 5d classic).

 

[qwerty]

Here is the 51k one.

 

[Tijn]

It took me a few minutes to do the raw conversion, and I had problems with a magenta cast to images, so hopefully I can save the rest of you a bit of time.

I got a similar "view" of the files using IrfanView. I am lead to believe that it is a faulty RAW conversion and that useful claims about noise cannot be made.
Look at a 600% enlargement of the left top corner of the ISO 6400 image to see what I mean. 8x8 pixel "blocks" with articulated edges are visible that introduce incorrect "detail" that will mess with any kind of noise-reduction one might want to perform on it. I can only assume that that is a RAW conversion error.