anti-aliasing

[jbills]

not running s+r yet at work here, but looking over the shoulder of a collegue, noticed that a shape leaning at a certain diagonal angle was showing a pretty bad case of aliasing. This is in a 2k workspace, which you would think would be fairly alias-free.

The workaround seemed to be to set the blur of the shape to .01. Actually, funny thing - I suggested to type in .025 because I read perry making that suggestion a while back (couldn't find the thread), and it immediately jerked the amount to .01, which seemed odd. But regardless, that gave a nice aliased edge.

Related - what scale is that blur amount on? Seems that a 1.0 equals 1 pixel scale would be most logical in terms of how most of us think. As it is, adding a blur of 1 completely blows out the edge. Are you all against absolute values? I know they are better for working with proxies, etc... but perhaps there's a better system we could adopt where working at 100% quality, a value of 1 equals 1 pixel.

back to aliasing - I haven't see this sort of aliasing before in any other programs - usually a smoother edge is highly desirable and seems to be the default. Can't think of an instance where we would want this sort of aliased edge. Perhaps all of the other programs are just applying some sort of transparent default blur. (which, that would suck, too, for obvious reasons)

Also, have noticed that on linux it's fairly cumbersome to type numbers into the fields because of the context-cursor sensitive nature of the Linux OS - perhaps it's time to add "virtual sliders" to the fields. where you can click in a field and drag to change the value, as in most modern interfaces of other software.

Thanks!

[paulm]

The blur scale (like all of our parameters) is resolution independent - it is not in pixels. A value of 100 would be "session height" pixels (which I admit is way too large).

As suggested elsewhere I will be adding a preference that lets you see/edit the blur values as pixels.

As for the aliasing, it is almost impossible to make a completely smooth edge on a mostly vertical/horizontal line with computer graphics/pixels. The edge algorithm isn't the best, but it's fast, and can be touched up with a bit of blur. I'll continue to look for better ones that don't undermine efficiency.

[jbills]

strange that it knocks an entry of .02 down to .01 - anything going on there?

[paulm]

Probably roundoff. The resolution of the field is only two decimals. I suspect your 0.02 is getting rounded to 0.0199999, which is then getting displayed as 0.01. Internally, it's probably still "almost 0.02".

It'll make more sense when I change it to show pixels.

[i8bozo]

wahoo! pixels.

[benedict_gs]

Has this problem been solved or is this still the current work around?

thanks

[marcop]

This was from quite a while ago on a very early version of Silhouette. Are you seeing aliasing problems with Silhouette v3?

[benedict_gs]

Yes we are on 3.0.4 and comps have been complaining that there is aliasing on the mattes. Is there a filter that we can adjust?

[marcop]

Just add a minute blur to your shapes. Their is a blur parameter for shapes. You can select multiple shapes in the Object List and adjust the blur for all selected shapes at the same time.

[paulm]

Though it would be better to use the Roto Node's global blur - it'll apply a uniform blur to everything all at once, and is much faster than applying blur to individual shapes.