Am 28.06.2021 um 02:17 schrieb Bald Eagle:
> _I_ can detect it at 100%, I don't think I'm completely imagining some sort of
> issue at 200%, and at 500%, that first "N" is an atrocity, and "f" is not
> looking good either...
My guess is that what you're unhappy with are just the artifacts
generated by `+am3`.
Anti-aliasing mode 3 is noisy by nature, as it uses pseudo-random jitter
across the entire pixel for supersampling.
Some of your other anti-aliasing settings exacerbate the problem:
`+r2` means you get at most 16 rays, which tends to be not a lot for
sharp contrasts. Using a higher value will not give you totally crisp
edges, but less noisy ones.
`+ag1` means you tend to get unnecessarily poor performance for
comparatively bright areas and unnecessarily poor quality in
comparatively dark ones. Note that `+agN` is NOT the space in which
averages are computed - that's always done in linear space - but rather
the space in which the threshold (`+a0.001`) is applied. A setting of 1,
i.e. linear gamma space, means the threshold allows for _physically_
identical maximum errors in regions of any brightness, but
_perceptually_ those maximum errors are more obvious in low-brightness
regions of the image than in high-brightness regions. Leaving it set to
the default of 2.5 causes POV-Ray to accept _perceptually_ similar
maximum errors regardless of absolute brightness. I strongly recommend
leaving it at the default unless you have very compelling reasons not to
`+acN` is missing from your settings entirely, although it is one of the
key settings for balancing performance vs. quality in mode 3.
`-j` doesn't hurt, but is completely ignored in mode 3. That mode is
jitter pure. (It may be argued that POV-Ray should possibly issue a
warning when faced with such a contradictory setting.)
(`+a0.001` is fine, no issue there. You could probably get away with a
higher value - `+a0.004` should be perfectly fine, maybe even `+a0.01`.)
(*A valid use case for `+ag1` would be if you tried to compute a depth
map rather than actual colors, and wanted the same absolute precision at
all distances within the range.)
>> As for the increase in thickness, it causes the text at the left and
>> right to appear more "bold" because we're seeing it stretch
>> perspectively "into the image", but it inevitably leaves the center
>> portion of the text as "unbold" as ever.
> Yes, I understand that - aside from scaling perpendicular to the viewing axis,
> there's likely not a viable way to bolden text aside from changing over to
> another font. But it's a sample scene, so I'm trying to stick with the 3 fonts
> that we ship with.
You could use two copies of the text object, and offset one of them
horizontally by a tiny value. That should bolden the vertical stems.
> Maybe I'll just scale x*1.something and see if that's good 'nuff.
You mean to shoo away the artifacts you're describing? It would be
surprising if it made any difference.
As for using this approach to bolden the text - it would widen the
vertical stems, but it wouldn't increase the overall "weight" of the font.
Post a reply to this message