Bad Areas Algorithm?

Hi,

I'm wondering if somebody can explain (or point me to a paper) exactly what is going on with the two "Bad areas artifacts" options?

I think I understand the general premise of a bad area: a block for which an accurate whole-block motion vector can't be found and that can't split into two moving regions (contour artifact).

Further I think I understand that the "suppression" setting turns off all interpolation for areas identified as "bad."

However, I'm not that clear on how the different masking types operate. (I understand masking in general, but sharp vs. light seems like an apples-to-oranges scale, and I'm not quite seeing how these masks would be defined.)

Regardless I don't have too many problems with artifacts, except with window blinds, which always seem to have really ugly artifacts.

Thanks for all the hard work. SVP is an amazing project and I've been happy to donate in the past. I'd love to see a "tech details" documentation section that really goes into algorithmic details, links papers, etc. I'm fairly familiar with video compression and graphics so I can grok most of what SVP is doing, but I'm sure I'm not the only one who's interested and curious to know more.

FDisk
I found that setting "Search Radius" to "1 pixel" solves this problem.

It is but by limiting radius to 1 pixel you also limit the ability to find fast motion and motion in opposite directions.
In the next version there'll be adaptive search radius based on local contrast which solves the problem in most cases.

FDisk
better set "Search Radius" to "2 pixels"?

Larger radius - more motion can be founded.
BTW, some discussion on doom9.org.