[argyllcms] Re: colprof quality
- From: Gerhard Fuernkranz <nospam456@xxxxxx>
- To: argyllcms@xxxxxxxxxxxxx
- Date: Mon, 3 Feb 2020 00:18:25 +0100
Am 02.02.20 um 23:30 schrieb graxx@xxxxxxxxxxxx:
Last but not least, if I want to compare predicted vs actual, I use colverify?
Meaning, I have to create another set of data points from which the profile
performance can be compared? Through the profile transforms?
Exactly. A prepresentative accuracy estimation can only be done via validation
against test data which are different from the training set which was use to
fit the profile, or possibly via cross-validation, if a separate test set is
not an option.
[ Note, particularly with a CLUT profiles you can almost always achieve very
low self-fit errors by simply reducing the regularization/smoothness (thus
increasing the effective number of parameters). This does not mean, though,
that the resulting profile is good or better - rather the opposite will be the
case, if you validate against a different, independent test set. ]
Regards,
Gerhard
Other related posts: