Spatial gain in DNG OpcodeList2

[gravures]

hi,
i have a dp2 merrrill and use darktable for raw developement. The colors were very dull and i noticed DT didn't read colors matrix, so i hard coded a matrix in DT, great improvement.
After that, i implemented in rawspeed (the lib that actual read raw in DT) an OpcodeList2 handler for spatial gain correction. Seems to do the job but with over corrected result. By errors and trials i approximate weighting factors to dim the corrections (1.8 for red, 2.3 for green and 1.7 for blue). With this, it's look way better but not perfect at all.
Now i try to find what's involve in this over corrected spatial gain to handle it nicely and finish the job. With r,g and b not having the same over correction power i'm thinking of white balance stuff.
But i've no precise idea of what to do here, any ideas?
Thanks for your help

[rolkar]

Just so we are on the same page. I assume you are using (an unmodified) X3F Tools to generate a DNG file? And then you try to modify dark table to be able to handle it nicely.

Personally i would then try to read the DNG file with some Adobe tools, like Lightroom or Photoshop in order to get a reference what is possible to achieve. This for two reasons.

1. We do not really understan how X3F field flattening works. This is the main reason why we gave up.

2. Different sensors (before Quattro) vary wildly. What might look Ok with one camera is totally off with another one.

So, not to use time on something that is impossible, try it first with some adobe reader. If the Adobe reader manages to do a much better job, then you have a potential for improvement.

[gravures]

Thanks for your reply. Yes my X3F TOOLS is unmodified. Unfortunatly i'm on linux and don't have lightroom on the hand, but sure I'll like to know how adobe handle this dng.
In fact I'm very close to a perfect rendering, on some dng, color cast is almost invisible and the final image is better than with Sigma Photo Pro.
Thanks for your advise and the work you've done with this tool.