Weird MSmooth YV12 mask bug?

[Scintilla]

When doing random seeking through scripts using MSmooth (2.0.2) in YV12 mode with AVISynth 2.5.6 and parameters mask=true, chroma=true in VirtualDubMod (1.5.10.1), I'm seeing this very annoying bug on many frames where it looks like the red offset channel has been taken from a frame in a completely different part of the video:

http://s93771304.onlinehome.us/msmoothbug.png

At first I thought this was merely due to my using MPEG-2 streams with the first frame lost, but today I realized that it also happens when the video I'm serving is, say, a HuffYUV AVI.

And no, setting chroma=false doesn't help.

MSharpen 1.10 doesn't have this problem, however.

Does anybody know how I can fix this? Aside from always running MSmooth in RGB32 mode, that is...

[foxyshadis]

Mask is luma-only, the chroma channel will always be random trash, which tends to come from a frame (sometimes the one you're looking at, usually not) and sometimes it's just green junk. Same effect occurs in masktools if you specify u=v=1, or "ignore". Add .greyscale to it if it really bothers you. It's not really anything to worry about, you either use only the luma channel in masks or use FitY2UV to extend it to chroma.

[mg262]

... or there's the new luma option which Manao kindly added:

http://forum.doom9.org/showthread.php?p=774117#post774117

[Scintilla]

Mask is luma-only, the chroma channel will always be random trash, which tends to come from a frame (sometimes the one you're looking at, usually not) and sometimes it's just green junk.
Really? I had thought the chroma made a difference in the masks, after seeing light blue edges show up in addition to the normal light green ones when I first switched to chroma=true (this is on the good frames, I mean).

Well, it's good to know that it's not screwing anything up. Thanks! :)

[foxyshadis]

Actually if it is supposed to create a good chroma mask, it's a bug... the bright points may be leakage from other (deallocated) frames. It'll just show whatever happened to be in memory, since it's faster than zeroing the planes, and if the chroma mask really is written it probably writes on top of that. I'd have to look at the source first to see which it is.

Edit: Nope, it only generates a mask for luma, and copies that to chroma. I think it does have a bug, because it treats the full garbage chroma plane as a mask, but all it would do is make the smoothing less effective and perhaps less stable than it should be.

[xorz]

To get rid of the chroma artifacts in msmooth mask mode, you can try this new version:
http://neuron2.net/board/viewtopic.php?t=847