On Mon, Sep 13, 2010 at 05:35:55PM +0930, David Gowers wrote: > I've just implemented > A) a better test for roundtripping (slow, though.) > > According to this test, we currently badly roundtrip in approximately > half of the 256 possible cases. Most but not all of these errors are > truncation errors.
I just checked it, I think it's 2% of the pixels, not 50%. Just remove pixels with alpha <= 1 from the input and the error goes away completely. We can't do anything about the roundtrip error at alpha <= 1 because after premultiplying the original 8bit value is squeezed into a range that contains fewer than 256 values. -- Martin Renold _______________________________________________ Mypaint-discuss mailing list [email protected] https://mail.gna.org/listinfo/mypaint-discuss
