Does this mean that until the PR is merged, even lossless compressed will not work?
Today I downloaded some sample images from mattgranger.com. Lighttable displayed them OK, but any attempt to read them in darkroom complained that it could not read the white balance information. I had assumed these were lossless compressed files, and exiftool shows:
File Type : NRW
File Type Extension : nrw
MIME Type : image/x-nikon-nrw
Exif Byte Order : Little-endian (Intel, II)
Make : NIKON CORPORATION
Camera Model Name : NIKON Z 9
I thought from that last line that this meant it was lossless compressed, but maybe the first line is showing otherwise (I have enquired of Matt Granger whether they are lossless or not - awaiting an answer). Since my own D810 files show:
File Type : NEF
File Type Extension : nef
MIME Type : image/x-nikon-nef
I guess I have downloaded HE* files. Is there a lossless compressed file available somewhere to download, so I can test my setup?
Thanks. I tried a couple, and got messages about unsupported input profile replaced with linear REC709.
The images were sort of visisble, so I guess my setup is ok.
This is not enough to identify whether this is “classic” lossless or HE, it just identifies a broad Nikon scheme. You have to take a look at the NEFCompression tag, which has changed location in Z 9 (but exiftool can figure it out anyway).
That image appears three times. With two of the downloads I get failed to read whitebalance messages. I assume they are the HE versions. With the first one I get the unsupported input profile message, and linear REC709 RGB is used, which results in a highly magenta cast.
Looking at my cameras.xml, and comparing it with your PR, I see I lack the ColourMatrices section. If I correct that, then it works fine.
There’s a bit of term overloading here. Just glancing at the dt code, I think the comment is targeting profile transforms in general, whereas “input profile” in the UI is AFAIK about the specific camera profile, for which REC709 should definitely not be used…
@danny’s addition to the adobe_coeff.c file (post #1 in the thread) seems to be appropriate for a Nikon camera; I used those numbers to process a Z 6 file I had handy and the colors retain the hues originally rendered by the Z 6 matrix. With those, no need to specify a separate input profile.
Just inspected my dt git clone, and yep, it disappeared from src/external. So, it looks like LibRaw is now the dt raw library (??), and the Libraw file with the adobe_coeffs is src/tables/colordata.cpp
For the last time: there is no more adobe_coeff.c since 4.0/master. The matrices now come directly from cameras.xml, like all other rawspeed info. The original post was written in 3.8 release time. I was very clear that my PR applies to post-3.8 master, i.e. now 4.0 and later.
And yes, for the last time: the Z 9 support is not officially merged in (at the time of this writing), so you will have to add it manually to cameras.xml every time you build until it is.
Nope. This is true only for Canon CR3 raw files (ATM of course). All other raws are still parsed by rawspeed.
@ggbutcher Do you have a better suggestion on what the input profile should fall back to if absolutely nothing is available?
LibRaw’s colordata.cpp and rawspeed’s cameras.xml have nothing to do with each other (apart from mining their data from the Adobe DNG Converter output).