Tweak 'interp' parameter

[Gavino]

In the Tweak() filter, the relationship between minSat, maxSat and interp is a bit confusing.

The doc (http://avisynth.org/mediawiki/Tweak) says that interpolation is done in the range [minSat-interp, minSat] and [maxSat, maxSat+interp]. However, the minSat and maxSat arguments are percentages, whereas in the implementation, interp is treated as an absolute saturation value and interpolation is actually done for saturations in the range [minSat'-interp, minSat'] and [maxSat', maxSat'+interp] where minSat' = minSat*1.19 and maxSat'=maxSat*1.19.

So, which is wrong - code or documentation?

Incidentally, the error message for an invalid interp value refers to P instead: "Tweak: P must be greater than or equal to 0 and less than 32."

[IanB]

Documentation updated.

Error message corrected.

[Gavino]

Now I'm more confused. :confused:

The updated documentation says interp is "given as a relative value".
I would take this to mean corresponding to saturation on a scale [0, 100], when in fact interp corresponds to saturation on the scale [0, 119] which I would call an absolute value.

[IanB]

Okay, relative is not the ideal term, I initially typed absolute but I didn't really like it because I was thinking it implied absolute U or V increments, which is still wrong.

Perhaps if we describe the implementation someone can come up with a better word.

For the purposes of the code in Tweak saturation is implemented as :- Sat=Sqrt((u-128)**2 + (v-128)**2)

Saturation adjustment, adjustSat', is linearly adjusted between adjustSat and 1.0 over the 2 ranges [minSat, minSat-Interp] and [maxSat, maxSat+Interp].

For the range minSat-Interp < Sat < maxSat+Interp, both the u and v values have the function applied :- f(x) = clamp((x-128)*adjustSat'+128, 16, 240)

[Gavino]

I've had a go at rewriting the description of interp in the hope of making it both clear and correct (difficult to do both at the same time :)). Anyone else is of course free to improve it if necessary.

I feel it would have been simpler and more consistent if interp was measured in the same units as minSat and maxSat, but we're stuck with that implementation now.

[IanB]

I'm not wedded to Interp continuing as it is. We can easily multiply it by 1.19 and call it a bug fix. And I don't really see why we currently need to limit it to 32, it could happily go up to 180.

Well I do know where the 32 came from. The initial implementation had Interp as an integer exponent of 2 between 0 and 5, default 4. Thus the current default of 16 and max of 32.

Thoughts :confused:

[Gavino]

I'm not wedded to Interp continuing as it is. We can easily multiply it by 1.19 and call it a bug fix. And I don't really see why we currently need to limit it to 32, it could happily go up to 180.
I'd be happy with that, as that's the way I originally thought it worked (before I looked at the code). The effect on backwards compatibility should be minimal.

In the new units, I suppose the extended upper limit would then become 150.