[argyllcms] Re: A few questions + an idea for spyd2.c (this time even more useless than before!)

  • From: Nikolay Pokhilchenko <nikolay_po@xxxxxxx>
  • To: argyllcms@xxxxxxxxxxxxx
  • Date: Wed, 11 Mar 2009 23:57:35 +0300

> Dear Nikolay!
> > You should check Your profile by the other set of points, than the
> > profile built. 
> Thank you for your suggestion.
> I do fully agree with the approach you proposed! Unfortunately I do not know 
> how I
> should deal with that... Do you know of any way of generating the
> testing points that would be different from 2000 points I used when
> generating the profile? I have the .ti1 and .ti3 files used to
> generate the profile.

For example:

1. You have successfully built the profile from 2000 points. It's OK. The name 
of profile "Profile.icc" for example.

2. You need to generate a test set.
The best of all, IMHO, will be uniform perceptual spread over Your device 
gamut, so targen is needed:

targen -v -d3 -f 100 -e 1 -c "Great.icm" -I -w Test_100

Now You have another "Test_100.ti1", which contain 100 test patches, most of 
them are unmatched with Your original 2000-patches of "Profile.ti1". It's Your 
validation set.

3. You have to measure the test set in the same conditions, as profile work:

dispread -v -yl -k "Profile.cal" Test_100

Now You have the measured validation data "Test_100.ti3"

4. You do the validation of Your profile:

profcheck -k -x -w -e Test_100.ti3 Profile.icc

Now Your 2000-point profile have checked by another 100 (92) test points! The 
result of profcheck reflects the real profile quality.

5. After all You can combine Profile.ti3 and Test_100.ti3 files into one 
2100.ti3 without any measurements. Just by adding 100-patches dataset after 
2000 patches of Your original *.ti3. Don't forget to change NUMBER_OF_SETS 
value to 2100 :) . If You do the 2100-points profiling again with the same 
parameters as initially, the new profile need not a new validation. It's errors 
will be within the validation result at 4-th point.

It's my workflow. Any comments, suggestions?

Other related posts: